CONFERENCE GUIDE

11th International Conference onControl, Automation, Robotics and Vision

(ICARCV 2010)

December 7-10, 2010

Grand Copthorne Waterfront Hotel Singapore

ICARCV 2010

11th International Conference onControl, Automation, Robotics and Vision

December 7-10, 2010Grand Copthorne Waterfront Hotel Singapore

Organized by

School of Electrical and Electronic EngineeringNanyang Technological University, Singapore

Technical Co-Sponsored by

IEEE Control Systems Society

IEEE Computational Intelligence Society

IEEE Robotics & Automation Society

Supported by Held in

TABLE OF CONTENTS

Welcome Messages............................................................................................................ 1

Committees....................................................................................................................... 4

Information about Singapore............................................................................................... 6

General Conference Information .......................................................................................... 8

Conference Venue ............................................................................................................10

Floor Plan of Conference Venue..........................................................................................11

Keynote Addresses............................................................................................................13

Panel Sessions..................................................................................................................16

Technical Sessions & Abstracts ...........................................................................................21

Reviewers ...................................................................................................................... 132

Author Index.................................................................................................................. 137

Program Overview .......................................................................................................... 148

1

MESSAGE FROM THE GENERAL CHAIR

Dear Friend and Colleagues,

On behalf of the organizing committee, we are pleased to welcome you to Singapore and to the11th International Conference on Control, Automation, Robotics and Vision (ICARCV 2010)!

The biennial event of ICARCV has proven to be a premium forum for researchers and engineersfrom all around the world to share their latest research results and accelerating the exploitation ofnew technologies in Control, Automation, Robotics and Computer Vision. The organizingcommittee is extremely delighted to have received overwhelming worldwide responses forICARCV 2010, and we are expected to host 400 delegates! The strong support from theTechnical Program Committee has enabled us to put together a very solid technical program forthe delegates. Besides the parallel technical sessions, there are also keynote addresses and paneldiscussions on the state-of-art development in Control, Automation, Robotics and ComputerVision to be delivered by eminent professors and researchers.

We are indeed honored to have Professor Minyue Fu from the University of Newcastle, Australia,Professor Bruno Siciliano from University of Naples Federico II, Italy and Professor ZhengyouZhang from Microsoft Research, USA as the keynote speakers for ICARCV 2010. Theirpresence would undoubtedly act prestige to the conference as they are the giants in theirrespective fields. I would like to express my sincere appreciation to all the keynote speakers andpanelists for their contributions and supports to ICARCV 2010.

I would like to thank all the organizers of invited sessions and the numerous researchersworldwide who helped to review the submitted papers. I am also grateful to the distinguishedInternational Advisory Committee members for their invaluable supports and assistances. Iwould like to gladly acknowledge the technical sponsorship provided by the IEEE ControlSystems Society, the IEEE Computational Intelligence Society and the IEEE Robotics andAutomation Society. I also wish to place my hearty thanks to all the members of the OrganizingCommittee for their hard work to make this conference possible, and to many friends, colleaguesand indeed family members who have helped the conference directly or indirectly. Last but notleast, I am grateful for the strong and enthusiastic support of all delegates including many oldfaces from around the world.

I hope that you will find your participation in ICARCV 2010 stimulating, rewarding, enjoyable andmemorable.

Changyun WENGeneral Chair of ICARCV 2010

2

MESSAGE FROM THE TECHNICAL PROGRAM CHAIR

Dear friends and colleagues,

On behalf of the Technical Program Committee, we welcome you to the 11th InternationalConference on Control, Automation, Robotics and Vision (ICARCV 2010). The conference is aforum for scientists, engineers and practitioners throughout the world to present the latestadvancement in Control, Automation, Robotics and Computer Vision. We have assembled an excitingTechnical Program covering all topics in these areas.

We received a very good response to the Call for Papers with a total of 899 submissionsincluding invited session papers from 56 countries/regions. All the submitted papers wereprocessed by the Technical Program Committee which has 95 members who are well knownresearchers with vast professional experience in various areas of the conference. Ten to fifteenpapers were assigned to each member. All the members worked professionally, responsibly anddiligently in soliciting expert international reviewers. Besides evaluations from reviewers, eachmember also provided his/her own assessments on the papers assigned, so as to ensure that onlyhigh quality papers would be accepted. After going through such a rigorous reviewing process,we finally included 329 papers for presentation in 59 oral sessions and 106 papers in 3poster sessions in the Technical Program. These papers contain the State-of-the-Art researchresults in both theory and applications in the areas of control, automation, robotics and computervision. In addition, the Technical Program also includes 3 excellent keynote addresses and 2 panelsessions. The 3 keynotes are

Audio-Visual Joint Processing for Active Object Detectionby Professor Zhengyou Zhang, Microsoft Research, USA09:15, Wednesday, December 8, 2010

Network based Control and Estimation Problemsby Professor Minyue Fu, The University of Newcastle, Australia08:30, Thursday, December 9, 2010

Robots Moving Closer to Humansby Professor Bruno Siciliano, University of Naples Federico II, Italy08.30, Friday, December 10, 2010

3

The details of the 2 plenary panel sessions are

Panel Session 1: Imaging and Vision: Challenges and Opportunities

Time: 13:30-15:30, Wednesday, 8 December 2010Panelists: Professor Zhi-Qiang Liu, City University of Hong Kong, SAR, China

Professor Brian Lovell, University of Queensland, Australia

Professor Massimo Tistarelli, University of Sassari, Italy

Professor Zhengyou Zhang, Microsoft Research (MSR), USA

Panel Session 2: Control and Autonomy: Challenges and Opportunities

Time: 13:10 - 15:10, Thursday, 9 December 2010Panelists: Professor Pedro Albertos, Polytechnic University of Valencia, Spain

Professor Patrick Doherty, Linkoping University, Sweden

Professor Minyue Fu, The University of Newcastle, Australia

Professor Daniela Rus, MIT, USAProfessor Bruno Siciliano, University of Naples Federico II, Italy

As in the past ICARCV conferences, we will present the Best Paper Award during theConference. Based on reviewers' comments and also the recommendation from TPCmembers, 14 papers were nominated for the best paper award. These 14 papers were furtherevaluated by some of our distinguished members of the International Advisory Committee toshortlist six finalists. The Best Paper will then be chosen by the Best Paper Selection Committeeafter further assessment of the oral presentations of the six finalists during the conference. Theselection of the best paper is based on technical quality and presentations including both writtenand oral presentations. The Best Paper Award will be announced and presented during theConference Banquet.

A CD-ROM containing preprints of all the papers scheduled in the program is provided at theconference to each registered participant as part of the registration package. The officialconference proceedings will be published by IEEE and included in the IEEE Xplore Database.

On behalf of the Technical Program Committee, we would like to thank all our reviewers for theirexpert reviews which are invaluable to the Committee in making a fair decision onacceptance/rejection of each paper. We are also grateful to the distinguished InternationalAdvisory Committee members for their invaluable support and assistance. Thanks also go to thededication, diligence and commitments of the Invited Session Chairs Professors Chien ChernCheah, Eric Sung and Lihua Xie and all the members of the Technical Program Committee. Wealso wish to place our hearty thanks to many friends, colleagues and indeed family members whohave helped completing the technical program directly or indirectly.

Thank you for your active participation of the conference. I hope that you will find yourparticipation in ICARCV 2010 in Singapore stimulating, rewarding, enjoyable andmemorable.

P. N. SuganthanTechnical Program Chairman, ICARCV 2010

4

COMMITTEES

Organising CommitteeGeneral Chair Changyun WENAdvisor Danwei WANGTechnical Program Chair P. N. SUGANTHANPanel and Keynote Chair Yeng Chai SOHInvited Session Chair Lihua XIEInvited Session Co-chairs Eric SUNG

Chien Chern CHEAHFinance Chair Han WANGPublicity Chair and Secretary Patricia J Y WONGPublication Chair Eric SUNGLocal Arrangement Chair Chin Seng CHUA

International Advisory CommitteeBaillieul, John (USA)Borne, Pierre E. (France)Bose, Bimal K. (USA)Cao, Xiren (HK, China)Cassandras, Christos G (USA)Chai, Tianyou (China)Chen, Guanrong (Ron) (HK, China)De Luca, Alessandro (Italy)De Silva, Clarence W (Canada)Er, Meng Hwa (Singapore)Goodwin, Graham C (Australia)Guo, Lei (China)Hill, David (Australia)Ikeuchi, Katsushi (Japan)Isidori, Alberto (Italy)Kam, Chan Hin (Singapore)

Kaynak, M. Okyay (Turkey)Krstic, Miroslav (USA)Kumar, P. R. (USA)Laugier, Christian (France)Lee, Kok Meng (USA)Lewis, Frank L (USA)Newcomb, Robert W (USA)Ortega, Romeo (France)Pal, Nikhil R (India)Parent, Michel (France)Polycarpou, Marios M (Cyprus)Siciliano, Bruno (Italy)Slotine, Jean-Jacques E (USA)Wang, Danwei (Singapore)Zurada, Jacek. M. (USA)

Technical Program CommitteeChairP. N. SUGANTHAN (Singapore)

MembersAhn, Hyo-Sung (Korea)Andreadis, Ioannis (Greece)Cai, Wenjian (Singapore)Chen, Tao (Singapore)Chen, Weihai (China)Chen, Xiaoqi (New Zealand)Chen, Yang Quan (United States)Cheng, Xiang (Singapore)Cho, Siu-Yeung David (Singapore)Choi, Han-Lim (Korea)Dhia Aldeen, Mohammad Abbas (Australia)Ding, Zhengtao (United Kingdom)Doherty, Patrick (Sweden)Fan, Huijin (China)

Fang, Debin (China)Gao, Huijun (China)George, Koshy (India)Gravdahl, Jan Tommy (Norway)Guo, Yi (United States)Hou, Saing Paul (Singapore)He, Bo (China)He, Ming-Yi (China)He, Xiangjian (Australia)Hong, Tzung-Pei (Taiwan)Hu, Patrick (New Zealand)Huang, Guangbin (Singapore)Huang, Hesheng (China)Hwang, Kao-Shing (Taiwan)

5

Hwang, Yan Chye (Singapore)Jia, Kevin Jiancheng (United States)Jiang, Zhong-Ping (United States)Julsereewong, Prasit (Thailand)Kariwala, Vinay Kumar (Singapore)Khan, Asad I (Australia)Khoo, Suiyang (Australia)Kodagoda, Sarath (Australia)Koh, Tong San (Singapore)Kumar, Ganesh (India)Lam, Kenneth K.M (Hong Kong)Lanusse, Patrick (France)Lee, Peng Hin (Singapore)Lee, Tae-Eog (Korea)Li, Ning (China)Li, Yangmin (Macau, China)Li, Yunhua (China)Lin, Zongli (United States)Ling, Keck Voon (Singapore)Liu, Chao (France)Liu, Yungang (China)Low, Kay Soon (Singapore)Ma, Xudong (China)Machado, J A Tenreiro (Portugal)Manic, Milos (United States)Mao, Kezhi (Singapore)Martinet, Pphilippe (France)Mckenna, Stephen (United Kingdom)Nagata, Fusaomi (Japan)Phat, Vu Ngoc (Vietnam)Prabu, D (India)Qin, Shi-Yin (China)Ramalingam, Soodamani (United Kingdom)Roopaei, Mehdi (Iran)

Sadeghian, Alireza (Canada)Sakagami, Norimitsu (Japan)Seow, Kiam Tian (Singapore)Shi, Ling (Hong Kong, China)Sluzek, Andrzej (Singapore)Song, David (China)Song, Wenbin (China)Sree, KIRAN (India)Su, Chun-Yi (Canada)Su, Weizhou (China)Sun, Dong (Hong Kong)Sung, Eric (Singapore)Tahara, Kenji (Japan)Takemura, Hiroshi (Japan)Tan, Tele (Australia)Toh, Kar Ann (Korea)Tsai, Ching-Chih (Taiwan)Ucinski, Dariusz (Poland)Wan, Shuai (China)Wang, Liuping (Australia)Wang, Qing-Guo (Singapore)Wang, X (China)Wang, Yanwu (China)Wei, Baoguo (China)Xia, Yuanqing (China)Xie, Lihua (Singapore)Yang, Guilin (Singapore)Yau, Wei Yun (Singapore)Yuen, Pong C (Hong Kong)Zhang, Chengjin (China)Zhang, Weihai (China)Zhou, Junhong (Singapore)Zou, Yun (China)

Technical & Logistics Support CommitteeTan, Peng ChyeLim, Puay ChyeYock, Ching FongTan, Chai SoonNg-Yap, Poh Geok, Pamela

Chia, Chiu KiatSow, Peck HengNeo, Kok ChuanZiauddin, Mohamed

Conference SecretariatTan, Peng ChyeNanyang Technological UniversitySchool of Electrical and Electronic EngineeringSingapore 639798Tel: (65) 6790 4220E-mail: icarcv@ntu.edu.sg

6

INFORMATION ABOUT SINGAPORE

Singapore is a bustling cosmopolitan city populated with high-rise buildings and landscape gardens.Brimming with a harmonious blend of culture, cuisine, arts and architecture, Singapore is a dynamiccity that’s rich in contrast and color. In fact, you can even say that Singapore embodies the finest ofboth East and West.

Located in Southeast Asia, Singapore has a land area of about 710 square kilometres, making her oneof the smallest countries in the world and the smallest in the region – hence the moniker “The LittleRed Dot”. Although small in size, Singapore commands an enormous presence in the world today withits free trade economy and highly efficient workforce. Also, her strategic location in the region hasenabled her to become a central sea port along major shipping routes.

PeopleAt present, Singapore’s population stands at about five million people, with English as the mainlanguage of instruction, and a mother tongue for each major ethnicity. Coming together as a societyand living in harmony, there are four major races – namely the Chinese (majority), Malay, Indian andEurasian.

The ethnic Chinese form 74.2% of the Singaporean population, with the country’s original inhabitants– the Malays, comprising of 13.4%. The Indians make up 9.2%, and Eurasians, Peranakans andothers making up a combined 3.2%. Singapore's many races live together in harmony, united by themotto, "Many races, one Singapore", and it is this ethnic diversity that is one of Singapore’sstrengths.

Each community offers a different perspective of life in Singapore in terms of culture, religion, foodand language.

LanguageThe four official languages in Singapore's constitution are English, Chinese, Malay and Tamil.However, in recognition of the status of the Malay people as the indigenous community in Singapore,the national language of the country is Bahasa Melayu, or the Malay Language.The presence of other languages, especially the varieties of Malay and Chinese, has obviously had aninfluence on the type of English that is used in Singapore. The influence is especially apparent ininformal English, an English-based creole that is commonly known as Singlish. A badge of identity formany Singaporeans, it represents a hybrid form of the language that includes words from Malay, aswell as Chinese and Indian languages.

Sight SeeingYou can experience Singapore’s rich historical heritage by visiting many of the national monuments,museums and memorials located around the city. On your trip here, remember to take a walk alongone of the many heritage trails and enjoy the sights and sounds at various cultural precincts, notablyChinatown, Little India and Kampong Glam or visit the well-known landmarks for a completeSingapore journey.

ShoppingIf you prefer the bright city lights and being amidst the hustle and bustle, then you’ll be delighted toknow that there are numerous shopping malls, museums, and dining and entertainment hotspots tochoose from. Get into the thick of the shopping action at the iconic Orchard Road stretch, or party thenight away at the Clarke Quay or Boat Quay areas, both of which offer a myriad selection of nightlifeactivities.

CuisineIt would not be at all surprising if the phrase “eat to your heart’s content” originated from New Asia -Singapore. The reason being, in Singapore’s multi-racial melting pot, all the various Asian cuisinescompete in the battle of the taste buds: Malay, Chinese, Indonesian, Peranakan, Indian, Thai,Japanese and Korean. Adding to this delicious confusion is Singapore’s vast array of hawker stallsand restaurants that range from global franchises to gourmet delis to fancy six-star settings. Andthere are also the seemingly endless food promotions and food tours that run throughout the year,

7

including the Singapore Food Festival in July. Some of the local food includes Chicken Rice, ChilliCrab, Indian/Chinese Rojak, Laksa, Mee Rebus, Nasi Lemak and Roti Prata.

WeatherSingapore has a tropical climate with relatively uniform temperature all year round. Temperatureranges from a low of 24ºC to a high of around 31ºC every day, and relative humidity is high. Rainfallusually takes the form of sudden showers and storms, which pass quite quickly, although moreprolonged periods of rain can be expected during the monsoon season from November to January.

TransportSingapore has one of the best mass rapid transit systems in the world and the proposed conventionvenue is well within its reach. The nearest MRT stations to the conference venue are the EsplanadeMRT (CC3) and Promenade MRT (CC4) along the Circle Line. Taxis are also readily available and arecharged by metered fare. Please note that various surcharges are applicable, including for ridesduring peak hours (07:00 – 09:30, 17:00 – 20:00 hrs) and for booking of taxis. Guided tours by taxiwith trained taxi drivers are also available at reasonable rates. In addition, if you choose to exploreSingapore by bus, there is an excellent network of air- conditioned buses.

More about SingaporeTo find out more about Singapore, please visit the website:http://www.yoursingapore.com/content/traveller/en/browse/see-and-do.html

HoursMost shops open from 10:00 to 22:00, including Sundays and Public Holidays. Some conveniencestores such as 7- Eleven open 24 hours a day.

CurrencyThe local currency is the Singapore dollar. S$1.00 is about US$0.72 or Euro 0.59.

TelephoneVisitors may purchase overseas calling cards for inexpensive overseas calls. These can be purchasedat convenience stores and post offices. Some useful telephone numbers are given below.

Police: 999Ambulance: 995Fire-Brigade: 995Cab booking –

Comfort and CityCab: 6552 1111TIBS and SMRT: 6555 8888Premier (Silvercab): 6363 6888

8

GENERAL CONFERENCE INFORMATION

Organizer School of Electrical and Electronic Engineering, Nanyang Technological University

Technical Co-Sponsors IEEE Control Systems Society IEEE Computational Intelligence Society IEEE Robotics and Automation Society,

Language

The conference and all its activities will be conducted in English.

RegistrationAll delegates are required to register upon arrival. A conference kit will be given which will contain acopy of the CD-ROM proceedings and the Conference Guide. Each delegate will be given a namebadge and it has to be worn at all times during the conference. Delegates/Accompanying Personswho have registered for the Conference Banquet will also be given tickets.

The registration hours are:Tuesday December 7 17:00 – 20:00 Level 2, Foyer of Waterfront BallroomWednesday December 8 08:00 – 16:00 Level 4, Foyer of Grand BallroomThursday December 9 08:00 – 14:00 Level 4, Foyer of Grand BallroomFriday December 10 08.00 – 12:00 Level 4, Foyer of Grand Ballroom

Registration FeeIEEE Members $950Non-members $1100

The Conference Registration Fee includes one welcome reception, six tea-breaks, three lunches, oneconference banquet and one copy of the CD-ROM Proceedings.

Additional banquet ticket can be purchased at SGD100 per person.

Oral Presentation1. Please note that the time allocated to each oral presentation is 20 minutes. 15 minutes

on Presentation and 5 minutes for Q&A.2. Surrogate presenters are expected to be sufficiently familiar with the material being

presented to answer questions from the audience and inform the Session Chair beforethe start of the session that he/she will be presenting on behalf of the co-author(s).

3. Microsoft PowerPoint slides or Acrobat PDF for presentation on a LCD video projector isrecommended.

4. All the presentation rooms will be equipped with a computer running Windows XPoperating system, installed with Microsoft Office 2003 compatible plus Acrobat Readerversion 7 compatible and a LCD video projector.

5. Authors are advised to prepare their presentation files in a format compatible with oneof the above applications, with the files stored in a CD-ROM or a USB flash drive. Please

9

ensure that there is no virus in the USB flash drive.6. For authors using Mac computers, please ensure that your presentation file is compatible

with that mentioned in Item 4 above.7. Presentation files should be up-loaded into the computer in the assigned

presentation room 10 minutes before the start of your scheduled session. There willbe an assistant in the room to help you.

Poster PresentationThe poster session will be held at the Foyer on Level 4.

1. Authors will be provided with one A0 size (1189 x 841 mm) poster stand.2. Authors may prepare their own presentation in one A0 size paper or a few A4 size papers

to be pasted on the poster stand.3. Authors may select their own design layout, colors and contents within the frame. However,

the poster must be in Portrait orientation.4. Authors are asked to report to the person in-charge at the Foyer by 8.15 am.5. Posters will be displayed until 6pm. However, the authors are expected to be present near

their display during the morning and afternoon tea- breaks.

Welcome ReceptionA Welcome Reception will be held on Tuesday, 7 December 7 between 6.00 to 8.00 pm at the Foyerof Waterfront Ballroom, Level 2, Grand Copthorne Waterfront Hotel.

Cocktail Reception cum Conference BanquetThe cocktail reception will be held on Thursday, December 9, 2010 at 7.00 pm in the Grand BallroomFoyer, Level 4, Grand Copthorne Waterfont Hotel. The banquet will commence at 7.30 pm. Additionaltickets at S$100 can be purchased from the Secretariat.

Internet AccessComplimentary internet access will be provided at the Secretariat Room, Toucan Room, Level 4.Please limit your access to 15 minutes. In addition, wireless internet access will also be available atthe conference and access instructions will be provided during the registration.

Grand Copthorne Waterfront Hotel

392 Havelock Road Singapore 169663 ▪ T +65 6733 0880 ▪ F +65 6737 8880 www.grandcopthorne.com.sg Company Reg. No: 52898318X

� (Entrance via Kim Seng Road / Jiak Kim Street)

Kim Seng Road (Grand Copthorne Waterfront Hotel) Bus No From to 5 Pasir Ris MRT, Newton MRT Tiong Bahru MRT, Bukit Merah MRT 16 Raffles City, Orchard MRT Tiong Bahru MRT, Bukit Merah MRT 75 Holland Village, Botanic Garden Outram Park MRT, Shenton Way, CBD, Suntec City 175 Lavender MRT, Gelyang Lorong 1, Orchard MRT Clementi Interchange 195 Suntec City, Raffles City Tiong Bahru MRT 970 Holland Village Outram Park MRT

Zion Road (Food Centre) Bus No From to 5 Tiong Bahru MRT, Bukit Merah MRT Newton MRT , Pasir Ris MRT 16 Tiong Bahru MRT, Bukit Merah MRT Raffles City, Orchard MRT 75 Outram Park MRT, Shenton Way, CBD, Suntec City Holland Village, Botanic Garden 175 Clementi Interchange Orchard MRT, Lavender MRT, Gelyang Lorong 1 Terminal 195 Tiong Bahru MRT Suntec City, Raffles City 970 Outram Park MRT Holland Village

Havelock Road (Riverview Hotel) Bus No From to 51 Jurong East, Alexandra Chinatown, Bugis 64 Red Hill MRT Little India 123 Bukit Merah Orchard MRT 186 Queenstown MRT CBD 608 Bukit Merah CBD

Havelock Road (Copthorne King’s Hotel)

Bus No From to 51 Chinatown, Bugis Jurong East, Alexandra 64 Little India Red Hill MRT 123 Orchard MRT Bukit Merah 186 CBD Queenstown MRT 608 CBD Bukit Merah

1

3

2

4

1

3

4

2

11

FLOOR PLAN OF CONFERENCE VENUE

LEVEL 2 FUNCTION ROOM PLAN

Pre-Conference Registration at Foyer of Waterfront Ballroom on 7th December Welcome Reception at Foyer of Waterfront Ballroom on 7th December Lunches on 8th to 10th of December at Waterfront Ballroom

12

FLOOR PLAN OF CONFERENCE VENUE

LEVEL 4 FUNCTION ROOM PLAN

Opening Ceremony, Keynote Sessions, Panel Sessions at Grand Ballroom Oral Sessions at Grand Ballroom, Canary I/II, Kingfisher, Nightingale, Pelican,

Oriole and Bluebird Tea-breaks, Interactive Sessions at Foyer Secretariat Room at Toucan Room

13

KEYNOTE ADDRESS I

Audio-Visual Joint Processing for Active Object Detection

Professor Zhengyou ZHANGMicrosoft ResearchUSA

Wednesday, December 8, 201009:15 – 10:15Grand Ballroom, Level 4

Abstract: Active object detection is very important in many applications including robot navigation.Interaction with the dynamic surrounding environment by a human being or a robot is inherentlymultimodal, and two dominant modalities are, without question, audio and visual. In this keynote, Iwill describe some of our recent activities in audio-visual processing for active object detection andshow that multimodal fusion provides significant improvement in robustness and accuracy over thecase when audio and video are processed independently. Applications include: presence detection inan office environment using multi-sensory inputs from streams of video, audio, and mouse andkeyboard events; active speaker detection in a meeting room using microphone arrays and cameraarrays; speaker recognition from face images and audio signals.

Biography: Zhengyou Zhang is a Principal Researcher at Microsoft Research (MSR), the researchcenter of Microsoft Corp. and is a Fellow of the Institute of Electrical and Electronic Engineers (IEEE).He is the Founding Editor-in-Chief of the newly established IEEE Transactions on Autonomous MentalDevelopment (IEEE T-AMD), and is on the Editorial Board of the International Journal of ComputerVision (IJCV), the IEEE Transactions on Multimedia (IEEE T-MM), the International Journal of PatternRecognition and Artificial Intelligence (IJPRAI), and the Machine Vision and Applications. He was onthe Editorial Board of the IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE T-PAMI) from 1999 to 2004. He is listed in Who's Who in the World, Who's Who in America and Who'sWho in Science and Engineering.

14

KEYNOTE ADDRESS II

Network based Control and Estimation Problems

Professor Minyue FUUniversity of Newcastle, CallaghanAustralia

Thursday, December 9, 201008:30 – 09:30Grand Ballroom, Level 4

Abstract: Rapid advances in communications, networking and sensing technologies have enabledwide deployment of networked control systems, with applications ranging from automated trafficnetworks to mobile sensor networks and unmanned airborne vehicles. This is an area of research withmany new theoretical and technological challenges and requires multi-disciplinary collaborations. Thistalk will provide an overview of this emerging area and highlight some key research opportunities andchallenges from the control point of view. We will start with motivational examples in wirelessindustrial control networks and smart power networks and study the associated control andestimation problems in these two applications. We then discuss several new theoretical results onnetworked control and estimation, including quantized LQG control, quantized state estimation, stateestimation with packet dropouts; network-based consensus and network-based system identification.

Biography: Professor Minyue Fu received his Bachelor's Degree in Electrical Engineering from theUniversity of Science and Technology of China, Hefei, China, in 1982, and M.S. and Ph.D. degrees inElectrical Engineering from the University of Wisconsin-Madison in 1983 and 1987, respectively. From1987 to 1989, he served as an Assistant Professor in the Department of Electrical and ComputerEngineering, Wayne State University, Detroit, Michigan. He joined the Department of Electrical andComputer Engineering, the University of Newcastle, Australia, in 1989. Currently, he is a ChairProfessor in Electrical Engineering. In addition, he was a Visiting Associate Professor at University ofIowa in 1995-1996, and a Senior Fellow/Visiting Professor at Nanyang Technological University,Singapore, 2002. He holds a ChangJiang Visiting Professorship at Shandong University and a Qian-renProfessorship at Zhejiang University in China.

He is a Fellow of IEEE. His main research interests include control systems, signal processing andcommunications. He has been an Associate Editor for the IEEE Transactions on Automatic Control,Automatica and Journal of Optimization and Engineering.

15

KEYNOTE ADDRESS III

Robots Moving Closer to Humans

Professor Bruno SICILIANOUniversity of Naples Federico IIItaly

Friday, December 10, 201008:30 – 09:30Grand Ballroom, Level 4

Abstract: Robots! Robots on Mars and in oceans, in hospitals and homes, in factories and schools,robots fighting fires, making goods and products, saving time and lives. Robots today are making aconsiderable impact on many aspects of modern life, from manufacturing to healthcare. Reaching forthe human frontier, robotics is also vigorously engaged in the growing challenges of new emergingdomains. Interacting, exploring, and working with humans, the new generation of robots willincreasingly touch people’s lives. Unlike the industrial robotics domain where the workspace ofmachines and humans can be segmented, applications of intelligent machines that work in contactwith humans are increasing, which involve e.g. haptic interfaces and teleoperators, cooperativematerial-handling, power extenders and such high-volume markets as rehabilitation, physical training,entertainment. In this context, it is customary to distinguish between Cognitive Human-RobotInteraction (cHRI) and Physical Human-Robot Interaction (pHRI). This talk is aimed at discussing anumber of issues in pHRI concerning with safety, dependability and dexterity. The presentation willbe accompanied by videos illustrating experimental tests on both conventional and new lightweightrobots endowed with force and vision sensors.

Biography: Bruno Siciliano is Professor of Control and Robotics, and Director of the PRISMA Lab inthe Department of Computer and Systems Engineering at University of Naples Federico II. Hisresearch interests include force and visual control, human-robot interaction and service robotics. Hehas co-authored 7 books, 70 journal papers, 170 conference papers and book chapters. He hasdelivered 80 invited lectures and seminars at institutions worldwide. He is a Fellow of IEEE, ASME andIFAC. He is Co-Editor of the Springer Tracts in Advanced Robotics, and of the Springer Handbook ofRobotics, which recently received the PROSE Award for Excellence in Physical Sciences & Mathematicsand was also the winner in the category Engineering & Technology. Professor Siciliano is the Past-President of the IEEE Robotics and Automation Society.

16

PANEL SESSION I

Imaging and Vision: Challenges and Opportunities

Time: 13:30 – 15:30, Wednesday, December 8, 2010Venue: Grand Ball RoomChair: Professor Kar-Ann TOH, Yonsei University, Korea

Panelists: Professor Zhi-Qiang LIU, City University of Hong Kong, SAR, ChinaProfessor Brian C. LOVELL, University of Queensland, AustraliaProfessor Massimo TISTARELLI, University of Sassari, ItalyProfessor Zhengyou ZHANG, Microsoft Research (MSR), USA

ICARCV 2010 proudly presents a plenary panel session on ‘Imaging and Vision: Challenges andOpportunities’. We are honored to be able to host four prominent professors in the fields of imagingand vision to be the panelists. The objective of the plenary panel session is to provide an opportunityfor researchers, especially early career researchers, to interact with world renowned experts inimaging and vision and seek their views on current and possible future developments of these fields.During the session, panel members will share their vast experiences and visions with audiencethrough effective face-to-face dialogues.

The panel consists of four world class researchers and educators. They are:

Professor Liu, Zhi-Qiang has worked in both academic and industry for over30 years. He has taught in three Universities in Canada, the University ofMelbourne, Australia, and City University of Hong Kong, SAR, China. He iscurrently with the School of Creative Media, City University of Hong Kong. Hehas taught many courses such as Art and Technology, Digital Media, ComputerGraphics, Animation with Blender; Computer Architecture, Computer Networks,Artificial Intelligence, C programming language, Machine Learning, and PatternRecognition and supervised many software development projects in media artand media technology.

Professor Liu was a Senior/Principal systems engineer in electronics and communications industry inChina and Canada. He led the following major research and development projects: China's firstgeneration of inertia-based guidance systems for navigation of missiles and aircraft; Automatedmanufacturing systems; Channel equalization schemes in wireless and mobile communications; CDMAcoding scheme initiatives (written in C programming language), which was supported in part by theCanadian Government; Wireless data communications, in particular, for mobile office scenarios; andInternet/Intranet wireless data communications in enterprises. Professor Liu has received a numberof prestigious fellowships, scholarships. He is a recipient of Honorable Mention from the InternationalPattern Recognition Society for an Outstanding Contribution and the Outstanding Paper Award at theFirst International Conference on Machine Learning and Cybernetics cosponsored by IEEE Systems,Man and Cybernetics Society, and was a Distinguished Professor invited by the Department ofComputing Science, University of Alberta, Canada. Since 2005, he has been the International ChairProfessor at the Department of Computer Science, Tsinghua University, Beijing, CHINA andDistinguished Professor at the prestigious National University of Defense Technology, CHINA. He wasan IEEE Distinguished Lecturer in 2006. He is the Vice President of the International Consortium onUncertainty Theory. Professor Liu is the Conference Chair of ACM SIGGRAPH Asia 2011 to be held inDecember 2011, Hong Kong.

However, in order not to be swayed by vanity and to focus on academic matters and be able toexpress freely as a teacher, in principle Professor Liu doesn't join any professional associations, hencesaving the dues and time, but offers his service when needed by the community. Professor Liu hasbeen a Panel member in Research Grants Council, UGC Hong Kong for seven years. He serves on theboards of Hong Kong Innovation and Technology Commission (ITC), Hong Kong Cyberport IncueTrainAdvisory Committee and IncueTrain Vetting Committee.

17

Self-Validated Labeling of Markov Random Fields in Computer Vision by Professor LiuIn computer vision a large number of problems is about finding boundaries in a scene which areusually ill-defined due to lack of resolution, noise and occlusions, etc. Traditional approaches such asregularization and the well-known Laplacian of Gaussian (LoG) type filters throughout the 70s to early90s have not led to satisfactory results. We have found, however, combining with unsupervisedlearning, more specifically, clustering, with the theory of Markov Random Field (MRF) we are able toachieve marked improvements over some of the major techniques recently reported in theliterature. In this talk I will address in particular the problem of self-validated labeling of Markovrandom fields (MRFs), namely to optimize an MRF with unknown number of labels. We presentgraduated graph cuts (GGC), a new technique that extends the binary s-t graph cut for self-validatedlabeling. Specifically, we use the split-and merge strategy to decompose the complex problem to aseries of tractable subproblems. In terms of Gibbs energy minimization, a suboptimal labelingis gradually obtained based upon a set of cluster-level operations. By using different optimizationstructures, we propose three practical algorithms: tree-structured graph cuts (TSGC), net- structuredgraph cuts (NSGC) and hierarchical graph cuts (HGC). In contrast to previous methods, the proposedalgorithms can automatically determine the number of labels, properly balance the labeling accuracy,spatial coherence and the labeling cost (i.e., the number of labels), and are computationallyefficient, independent to initialization and able to converge to good local minima. We apply theproposed algorithms to natural image segmentation. Experimental results show that our algorithmsproduce generally feasible segmentations for Benchmark datasets, and outperform alternativemethods in terms of robustness to noise, speed and preservation of soft boundaries.

Professor Brian C. Lovell was born in Brisbane, Australia in 1960. Hereceived the BE in electrical engineering (Honours I) in 1982, the BSc incomputer science in 1983, and the PhD in signal processing in 1991: all fromthe University of Queensland (UQ). Professor Lovell is Project Leader of theAdvanced Surveillance Group in NICTA and Research Leader of the Security andSurveillance Group in the School of ITEE, UQ. He served as President of theInternational Association of Pattern Recognition 2008-2010, and is a SeniorMember of the IEEE, Fellow of the IEAust, and voting member for Australia onthe Governing Board of the International Association for Pattern Recognitionsince 1998. Professor Lovell was Program Co-Chair of ICPR2008 in Tampa,

Florida, and is General Co-Chair of ACPR2011 in Beijing, and ICIP2013 in Melbourne. The AdvancedSurveillance Group works with port, rail, and airport organizations as well as several national andinternational agencies to identify and develop technology-based solutions to address real operationaland security concerns.

Advanced Surveillance to Protect Critical Infrastructure, by Professor LovellIn this talk recent trends in modern surveillance networks will be discussed including topics such assuper-megapixel cameras, post incident digital PTZ, integration and fusion of video and non-videosensors, and multimodal remote biometrics including face and iris on the move. Current projectsunderway with port, rail, and local councils will be outlined in brief. Finally there will be videodemonstrations of NICTA low-resolution face search technologies.

Professor Massimo Tistarelli was born on November 11, 1962 in Genoa, Italy.He received the Phd in Computer Science and Robotics in 1991 from theUniversity of Genoa, Massimo Tistarelli is currently Full Professor in ComputerScience and director of the Computer Vision Laboratory at the University ofSassari, Italy. Since 1986 he has been involved as project coordinator and taskmanager in several projects on computer vision and biometrics funded by theEuropean Community. He is a founding member of the Biosecure Foundation,which includes all major European research centers working in biometrics.

His main research interests cover biological and artificial vision, biometrics,robotic navigation and visuo-motor coordination. He is co-author of more than 100 papers in scientificconferences, books and international journals. Prof. Tistarelli is the principal editor for the Springerbook “Handbook of Remote Biometrics”, published in June 2009. He holds an international patent foran automatic biometric system based on face modality (Int.l Patent G06K9/62 EU Patent G06K9/00F3;

18

G06K9/62C2S). He also firstly proposed a unique, advanced method for the integration of face andfingerprint modalities already at the feature level. In 1991 he was awarded the best paper awardfrom IEEE Computer Society.

Since 1998 he was the chairman for several International scientific events on biometrics. He was thechairman of the 5th IEEE AutoId conference, track chair for biometrics at the 19th ICPR, Area chairfor CVPR 2010 and general chair of the IEEE/IAPR 3rd Int.l Conference on Biometrics hold in June2009. Prof. Tistarelli is associate editor for the journals IEEE Transactions on Pattern Analysis andMachine Intelligence, Image and Vision Computing and Pattern Recognition. Since 2003 he is thefounding director for the Int.l Summer School on Biometrics (now at the 8th edition). He is the chairof the IAPR Technical Committee 4 on biometrics, President of the IEEE Biometrics chapter, Memberof the IEEE Biometrics Professional Certification Committee, Fellow member of IAPR and Seniormember of IEEE.

The state of the art, opportunities and challenges in face-based biometricsEnrolling to a graduation course and crossing the Singapore border may seem two very differenttasks. Yet they share a fundamental common operation: the verification of the user identity. Due to avariety of reasons ranging from the need of an increased security to the need to facilitate the accessto e-services, the process of identification requires a fast, easy and secure process. Among theavailable identification technologies biometrics, and in particular face recognition, has grabbed theattention of several sectors in our society.Despite the sometimes good or bad advertisement of these technologies, the identification of people’sfaces is a reality which is now possible thanks to a number of technologies, first of all ComputerVision and Image Understanding. The human face may seem to be a simple object to be recognized.After all we learn to recognize faces since birth. Yet, faces are quite complex objects: they have nofixed geometry, they do change their shape over time, the colour is not constant, the faceappearance may change due to different pose and illumination. These, as well as other peculiaritiesof human faces are still a real challenge for many computer vision and machine learning algorithmsapplied for the identification of human faces. The fact that several face recognition algorithms havebeen devised and successfully tested, and a number of commercial face recognition systems are nowon the market, is a proof that this is a mature and very promising technology. Moreover, thisdemonstrates the efficacy and success of Computer Vision research pursued over the past 40 years.Faces are intrinsically 3D objects. As a consequence, the analysis of 2D face images to infer usefulinformation about the imaged subject is an ill-posed problem. It can be overcome either by fusingmultiple images (to indirectly infer the missing information due to the 3D to 2D transformation) or bydirectly measuring the 3D face shape. Either using 2D or 3D information can lead to interestingconsiderations about the face structure. Moreover, 2D (textural) and 3D (geometrical) data can becombined to improve the identification accuracy.Several applications domain demonstrate how promising this technology is. From forensic practice toe-government, the analysis and identification of human faces constitutes a real challenge whichcurrent technologies based on Computer Vision research are ready to face.

Professor Zhengyou Zhang is a Principal Researcher at Microsoft Research(MSR), the research center of Microsoft Corp. and is a Fellow of the Institute ofElectrical and Electronic Engineers (IEEE). He is the Founding Editor-in-Chief ofthe newly established IEEE Transactions on Autonomous Mental Development(IEEE T-AMD), and is on the Editorial Board of the International Journal ofComputer Vision (IJCV), the IEEE Transactions on Multimedia (IEEE T-MM), theInternational Journal of Pattern Recognition and Artificial Intelligence (IJPRAI),and the Machine Vision and Applications. He was on the Editorial Board of theIEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE T-PAMI)from 1999 to 2004. He is listed in Who's Who in the World, Who's Who in

America and Who's Who in Science and Engineering. http://research.microsoft.com/~zhang/.

Opportunities and challenges of depth camera, by Professor ZhangWith the launch of Microsoft Kinect sensors for Xbox 360, depth cameras are expected to becomeaffordable for the vision and robotics communities because the mass game market will drive the costdown. This could present a revolution in our research. I will talk about the opportunities andchallenges accompanying the wide availability of depth cameras.

19

PANEL SESSION II

Control and Autonomy: Challenges and Opportunities

Time: 13:10 – 15:10, December 9, 2010Venue: Grand Ball Room, Level 4Chair: Professor Yeng Chai SOH, Nanyang Technological University, Singapore

Panelists: Professor Pedro ALBERTOS, Polytechnic University of Valencia, SpainProfessor Patrick DOHERTY, Linkoping University, SwedenProfessor Minyue FU, The University of Newcastle, AustraliaProfessor Daniela RUS, MIT, USAProfessor Bruno SICILIANO, University of Naples Federico II, Italy

ICARCV 2010 proudly presents the plenary panel session on Control and Autonomy: Challenges andOpportunities. We are honored to be able to invite five prominent professors in the field of controland robotics to be the panelists. The objective of the plenary panel session is to provide anopportunity for researchers, especially young researchers, to interact with world renowned experts incontrol and autonomy and seek their views on current and possible future developments of the fields.During the session, panel members will share their vast experiences and visions with audiencethrough effective face-to-face dialogues.

The panel consists of five world class researchers and educators. They include:

Professor Pedro Albertos, past president of IFAC (the International Federationof Automatic Control) in 1999-2002, and Senior Member of IEEE, is a worldrecognized expert in real-time control, leading several projects in the field. FullProfessor since 1975, he is currently at Systems Engineering and Control Dept.UPV, Spain. He is Doctor Honoris-Causa from Oulu University (Finland) andBucharest Polytechnic (Rumania). Invited Professor in more than 20 Universities,he delivered seminars in more than 30 universities and research centres.Authored over 300 papers, book chapters and congress communications, co-editor of 7 books and co-author of “Multivariable Control Systems” (Springer2004) and “Feedback and Control for Everyone”(Springer 2010), he is also

associated editor of Control Engineering Practice and Automatica and Editor in Chief of the Spanishjournal RIAI. His research interest includes multivariable control and nonconventional samplingcontrol systems, with focus on time delays and multirate sampling patterns.

Professor Patrick Doherty is a Professor of Computer Science at theDepartment of Computer and Information Sciences (IDA), Linkoping University,Sweden. He heads the Artificial Intelligence and Integrated Computer SystemsDivision at IDA. He serves as director of LinkLab, a research center for futureaviation systems, which is a collaborative endeavor between LinkopingUniversity and Saab Aero Systems. He is an ECCAI fellow and currently anECCAI board member. He has previously served as president of SAIS, theSwedish Artificial Intelligence Society. He is an associate editor for the ArtificialIntelligence Journal and is on the board of directors for KR Inc. His main areasof interest are knowledge representation, automated planning, intelligent

autonomous systems and multi-agent systems. A major area of application is with UnmannedAircraft Systems. His research group has successfully designed and deployed many autonomous UAVsystems in the past decade have also won several international competitions pertaining to micro-aerial vehicles and to automated planning. He has over 150 refereed scientific publications in hisareas of expertise and has given many keynote and invited talks at leading international conferences.He is also CEO of a new start-up company, UAS Technologies Sweden AB, which designs and marketsMicro-Aerial Vehicles.

20

Professor Minyue Fu received his Bachelor's Degree in Electrical Engineeringfrom the University of Science and Technology of China, Hefei, China, in 1982,and M.S. and Ph.D. degrees in Electrical Engineering from the University ofWisconsin-Madison in 1983 and 1987, respectively. From 1987 to 1989, heserved as an Assistant Professor in the Department of Electrical and ComputerEngineering, Wayne State University, Detroit, Michigan. He joined theDepartment of Electrical and Computer Engineering, the University of Newcastle,Australia, in 1989. Currently, he is a Chair Professor in Electrical Engineering. Inaddition, he was a Visiting Associate Professor at University of Iowa in 1995-1996, and a Senior Fellow/Visiting Professor at Nanyang Technological

University, Singapore, 2002. He holds a ChangJiang Visiting Professorship at Shandong University anda Qian-ren Professorship at Zhejiang University in China.

He is a Fellow of IEEE. His main research interests include control systems, signal processing andcommunications. He has been an Associate Editor for the IEEE Transactions on Automatic Control,Automatica and Journal of Optimization and Engineering.

Professor Daniela Rus is a Professor of Electrical Engineering and ComputerScience, where she is associate director of MIT's Computer Science and ArtificialIntelligence Lab (CSAIL) and co-directs the MIT Center for Robotics at CSAIL.Her research interests include distributed robotics and mobile computing andher application focus includes transportation, security, environmental modelingand monitoring, underwater exploration, and agriculture.

Rus is notable for spear-heading research in programmable matter bydeveloping the several self-configuring robots. In addition, she worked with herstudents to has designed, control, and field autonomous underwater robots,

agricultural robots that herd cattle, low-cost, early warning sensors for disaster prevention indeveloping countries, and teams of autonomour aerial vehicles that can monitor adaptively largeenvironments.

Rus is the recipient of the NSF Career Award and an Alfred P. Sloan Foundation Fellow. She is a Classof 2002 MacArthur Fellow and a fellow of AAAI and IEEE. Before receiving her appointment at MIT,Rus was a professor in the Computer Science Department at Dartmouth, where she founded anddirected two laboratories in robotics and mobile computing.

Rus earned her PhD in Computer Science from Cornell University.

Professor Bruno SICILIANO is Professor of Control and Robotics, andDirector of the PRISMA Lab in the Department of Computer and SystemsEngineering at University of Naples Federico II. His research interests includeforce and visual control, human-robot interaction and service robotics. He hasco-authored 7 books, 70 journal papers, 170 conference papers and bookchapters. He has delivered 80 invited lectures and seminars at institutionsworldwide. He is a Fellow of IEEE, ASME and IFAC. He is Co-Editor of theSpringer Tracts in Advanced Robotics, and of the Springer Handbook of Robotics,which recently received the PROSE Award for Excellence in Physical Sciences &Mathematics and was also the winner in the category Engineering & Technology.

Professor Siciliano is the Past-President of the IEEE Robotics and Automation Society.

21

Detailed Technical Session Listing and Abstracts

22

Session WA1 - Invited SessionDate : Wednesday, 08 December 2010Time : 1040 - 1200Venue : Ball Room

Advances in Biometric Theory andApplications IChairs : Wei Yun YAU

Institute for Infocomm Research

WA1.1 P1143 1040 - 1100Sparse Representations and RandomProjections for Robust and CancelableBiometricsVishal M PATEL, Rama CHELLAPPA, *MassimoTISTARELLIUniversity of Maryland*University of Sassari

In recent years, the theories of SparseRepresentation (SR) and Compressed Sensing (CS)have emerged as powerful tools for efficientlyprocessing data in non-traditional ways. An area ofpromise for these theories is biometric identification.In this paper, we review the role of sparserepresentation and CS for efficient biometricidentification. Algorithms to perform identificationfrom face and iris data are reviewed. By applyingRandom Projections it is possible to purposively hidethe biometric data within a template. This procedurecan be effectively employed for securing andprotecting personal biometric data against theft.Some of the most compelling challenges and issuesthat confront research in biometrics using sparserepresentations and CS are also addressed.

WA1.2 P1124 1100 - 1120Generating Cancelable Biometric Templatesusing a Projection LineTohari AHMAD, Jiankun HURMIT University

Biometric authentication systems can address theproblems of genuine user verification and usabilitysuffered by the conventional cryptosystems based onthe password and token. However, as biometrics isnot replaceable, its privacy has become a concern.Also, if the same biometrics is used in multipleapplications, it will be vulnerable to potential cross-matching attacks. In this paper, a projection linemethod is proposed for cancelable fingerprinttemplate design. The main advantage of this methodis its simplicity. Because there exist many possibilitiesprojecting minutia into the projection line controlledby the whole range of possible projection lineproperties, the proposed approach offers non-invertible transformation and also possess goodrevocability property.

WA1.3 P1137 1120 - 1140A Pitfall in Fingerprint Features Extraction

Peng ZHANG, Cai LI, Jiankun HURMIT University

In fingerprint feature extraction, it is perceivedthat pixel-level image rotation transformation is alossless transformation process. In this paper,investigation has been conducted on analyzing theunderlying mechanisms of ?ngerprint image rotationprocessing and potential effect on the major features,mainly minutiae and singular point, of the rotationtransformed ?ngerprint. Qualitative and quantitativeanalysis have been provided based on the intensiveexperiments. It is observed that the informationintegrity of the original ?ngerprint image can besignificantly compromised by the image rotationtransformation process, which can cause noticeablesingular point change and produce non-negligiblenumber of fake minutiae. It is found that thequantization and interpolation process can changethe fingerprint features significantly though they maynot change the image visually.

WA1.4 P1146 1140 - 1200MCC: A Baseline Algorithm for FingerprintVerification in FVC-onGoingRaffaele CAPPELLI, Matteo FERRARA, DavideMALTONIUniversity of Bologna

This paper describes an improved version of theMCC fingerprint matching approach. An in-deptherror analysis allowed us to point out the weakestpoints of the original MCC and to design: i) a moreeffective minutiae pair selection and ii) a moredistortion-tolerant relaxation. The parameters of thenew version have been tuned over a new largerdataset and the final algorithm has been evaluatedon FVC-onGoing. The results show that MCCcompares favorably with some of the most accuratecommercial algorithms published in FVC-onGoing.

23

Session WA2Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Canary I

Neuro-Fuzzy Control IChairs : Sundararajan NARASIMHAN

Nanyang Technological UniversityXiaolin ZHANGTokyo Institute of Technology

WA2.1 P0332 1040 - 1100Globally Exponential Stability of DelayedNeural Networks with ImpulsesJin ZHOU, *Quanjun WU, Lan XIANG, **Gang ZHANGShanghai University*University of Electric Power, Shanghai**Hebei Normal University

The present paper is mainly concerned with theissues of global exponential stability in recurrentdelayed neural networks in the presence of impulsiveconnectivity between the neurons. By establishing anextended Halanay differential inequality on impulsivedelayed neural networks, some simple yet genericcriteria for global exponential stability of such neuralnetworks are derived analytically. Compared withsome existing works, the distinctive feature of thesecriteria is that it is not necessary to learn the prioriinformation about the stability of the correspondingneural networks without impulses, which means therecurrent delayed neural networks can be globallyexponentially stabilized by impulses even if thecorresponding neural networks without impulses maybe unstable or chaotic itself. Moreover, examples andsimulations are given to illustrate the practical natureof the novel results.

WA2.2 P0632 1100 - 1120Comparison of Different Subset SelectionAlgorithms for Learning Local Model Networkswith Higher Degree PolynomialsOliver BAENFER, Benjamin HARTMANN, OliverNELLESUniversity of Siegen

A comparison of three different subset selectionmethods in combination with a new learningalgorithm for nonlinear system identification withlocal models of higher polynomial degree is presentedin this paper. Usually the local models are linearlyparameterized and those parameters are typicallyestimated by some least squares approach. For theutilization of higher degree polynomials thisprocedure is no longer feasible since the amount ofparameters grows rapidly with the number of physicalinputs and the polynomial degree. Thus a newlearning strategy with the aid of subset selectionmethods is developed to estimate only the mostsignificant parameters. A forward selection methodwith orthogonal least squares, a stepwise regressionand a least angle regression method are used for

training different neural networks. A comparison ofthe trained networks shows the benefits of eachsubset selection method.

WA2.3 P0141 1120 - 1140Linear Discrete-time Systems with FuzzyParametric UncertaintyPetr HUSEKCzech Technical University in Prague

The paper deals with the problem of determinationof stability margin of uncertain linear discrete-timesystems with parameter uncertainty described byfuzzy numbers. Arbitrary membership functionsdescribing the uncertainty of coefficients of char-acteristic polynomial are considered. The presentedsolution is based on transformation of the originalproblem to Hurwitz stability test and generalization ofTsypkin-Polyak plot.

WA2.4 P0597 1140 - 1200A Cooperative Interaction Control Methodologyof a Pair Independent Control SystemXiaolin ZHANGTokyo Institute of Technology

Human eyes cannot easily focus on two separatetargets at the same time. The two eyes have theirown cooperative movements that can be separatedinto conjugate and ver-gence movements (relativemovements) which have differ-ent dynamiccharacteristics. Here, in order to achieve rea-listic eyemovements, first a binocular motor control sys-temmodel was proposed based on the neural pathwaysof the binocular motor system. The model has crosspathways between the two symmetrical eye fs owncontrol loops which have never been found in anyconventional robot control system. Consequently, themodification on this ex-traordinary feature modelresulted that a pair of interacting control systemsthrough the cross pathways can be sepa-ratelyanalyzed into two independent control systems basedon its behaviors, i.e. the conjugate movement controlsys-tem and the relative movement control system.The effec-tiveness of this principle, then, was verifiedin the experi-ment through the binocular robot eyebased on this pro-posed control system model. Theconstructed robot can op-erate several cooperativemovement characteristics accord-ing to human-eyemovements. For example, it cannot focus on twodifferent targets at the same time, even though thetargets have the same shape.

WA2.5 P0979 1200 - 1220An Intelligent Fault-Tolerant Satellite AttitudeControl System with-out HardwareRedundanciesSureshkumar CHANDRASEKAR, Sundaram SURESH,Sundararajan NARASIMHAN, NarayanaswamyNAGARAJANNanyang Technological University

An intelligent fault-tolerant control design based

24

on nonlinear adaptive control theory is proposed formicrosatellite attitude control system, withoutproviding for redundant actuators. An intelligentneural adaptive fault-tolerant control law, based onfeedback error learning scheme is developed to aidthe baseline controller. Using this control law, thecontrol torques are suitably adjusted to achieve thedesired target acquisition mode under a partialactuator failure in the roll direction. Simulationstudies have been carried-out to examine thecapabilities of the proposed fault-tolerant controlsystem under partial actuator failure in the roll-axis.Results clearly show that the proposed intelligentcontroller is robust and requires a minimal settlingtime during the desired target acquisition mode.

Session WA3 - Invited SessionDate : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Canary II

Multidimensional Systems Modeling andSignal ProcessingChairs : Yun ZOU

Nanjing University of Science andTechnology

WA3.1 P0689 1040 - 1100

Sufficient LMI Conditions for H∞ Static OutputFeedback Control of 2-D SystemsZhi-Yong FENG, Li XU, Yoshihisa ANAZAWAAkita Prefectural University

This paper investigates the problems of H∞ staticoutput feedback (SOF) control of two-dimensional (2-D) discrete systems described by the Roesser modeland Fornasini-Marchesini second model, respectively.By applying the 2-D Bounded Real Lemma combinedwith a slack variable technique, some sufficient linearmatrix inequality (LMI)-based conditions for theexistence of SOF controllers are established.Numerical examples are given to illustrate theeffectiveness of the proposed LMI-based designmethod.

WA3.2 P0706 1100 - 1120Hybrid Traffics Congestion Control based on 2-D Hurwitz-Schur StabilityPengxuan MAO, Yang XIAO, *Guangzhi QU, **SeokWOO, **Kiseon KIMBeijing Jiaotong University*Oakland University**Gwangju Institute of Science and Technology

Classical network fluid model and RED algorithmare based on TCP flows in internet network, and theyhave not considered the UDP flows' effects innetwork. However, in real work of the network, thenetwork link capacity is shared by the hybrid traffics:TCP flows and UDP flows, and UDP flows can occupythe original link capacity of TCP flow. Since there isno feedback control for UDP flows, the classicalnetwork fluid model and RED algorithm can notreflect and control the congestion of TCP/UDPnetworks. To solve the problem, we modify theclassical AQM router into multiple queues AQM routerwith classifier. We express the proposed TCP/UDPAQM router by a linear time-delay system model.Then, by utilizing the 2-D Laplace-Z transformtechnique, we derive some explicit conditions thatestablish the relationship between the controlparameter pmax and the network stability. This paperfirst proposes parameter pmax's stability bounds forTCP/UDP routers for congestion control based on 2-DHurwitz-Schur stability conditions. The simulationresults verify that the proposed stability condition cangain the effective congestion control.

25

WA3.3 P0727 1120 - 1140Stability Analysis for 2-D Discrete Systemswith Varying DelayShuxia YE, Weiqun WANG, Juan YAONanjing University of Science and Technology

This paper is concerned with stability analysis fortwo-dimensional (2-D) discrete systems described bythe Roesser models (RM) with varying delay in thestate. By utilizing the delay partitioning idea and theLyapunov method, a new stability criteria is proposedin terms of linear matrix inequalities (LMIs). Thisresult is delay-dependent and also dependent on thepartitioning size. A numerical example is given todemonstrate the effectiveness and the benefits of thepresented methods.

WA3.4 P1129 1140 - 12002-D IIR Filter Bank Design by an LMI ApproachHuiling XU, Zhiping LIN, Cishen ZHANG, AnamitraMAKURNanyang Technological University

This paper is concerned with the problem of thetwo-channel 2-D IIR filter bank design. Using asystematic optimization approach, which is solved interms of an LMI, the 2-D IIR synthesis filters aredesigned such that the systems are alias free andthen optimized to achieve approximate perfectreconstruction.

WA3.5 P0425 1200 - 1220Linear Models for Plasma Current Control inTokamak ReactorsAitor J. GARRIDO, Izaskun GARRIDO, Jose M GorettiSEVILLANO, Modesto AMUNDARAIN, Mikel ALBERDI,Oscar BARAMBONES, Manuel DE LA SENUniversity of the Basque Country

The control of plasma in nuclear fusion has beenrevealed as a promising application of ControlEngineering, with increasing interest in the controlcommunity during last years. In this paper it isdeveloped a control-oriented linear model for thecontrol of plasma current. For this purpose, it isprovided a summary of the background necessary todeal with control problems in tokamak-based nuclearfusion reactors as it is the case of the future ITERtokamak. Besides, it is also given a review of themost used simulators and plasma models, with theaim of providing an adequate background for controlengineers to derive their own control-oriented modelor to choose the appropriate existing one. Finally, theproposed plasma model performance is proven in acurrent drive profile trajectory tracking problem usinga modified anti-windup PID-based control scheme

Session WA4Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Kingfisher

Sensor Network SystemsChairs : Lihua XIE

Nanyang Technological UniversityBoon-Hee SOONG

Nanyang Technological University

WA4.1 P0985 1040 - 1100Palm Size Charging Platform with UniformWireless Power TransferYun YOU, Boon-Hee SOONG, SelvakumaranRAMACHANDRAN, *Wei LIUNanyang Technological University*Singapore Institute of Manufacturing Technology

In this paper, the design of wireless inductivecharger for portable electronic devices is present.There are two main parts have been included. Thefirst part is the exciting circuit in which to providesufficient input power to the primary chargingplatform. The exciting circuit is comprised byoscillator circuit and power amplifier circuit. Both ofthe two circuits have been simulated in software andverified by measurements. The second part is a twolayer near uniform field planar inductive platform(primary coil). The design of the charging platformconsists of two layers of square spirals overlappingeach other using the principle of superposition. Thestrong areas (peaks) of one layer are superpositiononto the weak areas (troughs) of the other layer. Inthis manner, a near uniform magnetic field suitablefor power and signal transfer is created.

WA4.2 P0994 1100 - 1120Distributed RSSI-Sharing for Two-WayRanging Base Station SelectionBertrand TAY, Wei LIU, *Hin Fei CHAI, *Peter HanJoo CHONGSingapore Institute of Manufacturing Technology*Nanyang Technological University

While Two-Way Ranging (TWR) is generatinginterest as a accurate and low-cost indoor localizationtechnique, most TWR research focuses primarily onalgorithms or error reduction, with little regard to afundamental system-level issue: deciding which BaseStations (BS) should perform the TWR. While mostresearchers acknowledge the need for a scanningphase to select BSs before the ranging phase iscarried out, this scanning phase has not beenexplored in detail. In that regard, we utilize ReceivedSignal Strength Indicator (RSSI) as the basis for BSselection. RSSI is a measure of the signal strength ofan incoming radio packet, and is found on virtuallyevery Radio Frequency (RF) transceiver, making itsuse almost free of additional hardware cost. WhileRSSI measurements are unreliable in indoor

26

environments and cannot be practically used forlocalization, we show that RSSI is a reliable indicatorfor the scanning phase and BS selection. Morespecifically, we develop a Distributed RSSI-Sharingtechnique to efficiently select BSs and create asimulation to test this technique. We implement andtest technique on actual hardware with good resultsand finally propose RSSI-smoothing and a BS-Reduction method to improve the performance.

WA4.3 P1135 1120 - 1140A Lower Bound on Expected Localization Errorin Wireless Sensor NetworkDI MA, Meng Joo ER, Hock Beng LIM, Bang WANGNanyang Technological University

Localization is an important research problem inWSNs (wireless sensor networks) and many WSNlocalization algorithms have been proposed in theliterature. For a single sensor node whose anchors’positions are known, the location estimation errorlower bound can be computed by using the CRLB(Cramer-Rao Lower Bound). However, it is stillunclear to the research community what thelocalization error lower bound is from a network pointof view. In this paper, we derive a lower bound ofthe expected localization error for a network whosesensor nodes and anchors are randomly distributedaccording to a Poisson point process. We show thatthe lower bound of the expected localization error fora network is a function of the anchor density and thevariance of anchor-to-sensor distance measurements.

WA4.4 P1144 1140 - 1200Sensor Deployment Strategy for Random FieldEstimation: One-Dimensional CaseYang WENG, *Lihua XIE, **Wendong XIAO, ***SenZHANGSichuan University*Nanyang Technological University**Institute for Infocomm Research***Singapore Polytechnic

Deploying the sensor nodes at the best locationsfor random field reconstruction via sensor network isa fundamental task. One-dimensional random field isa stochastic process. In this paper, we first proposean optimal sensor deployment strategy for Wienerprocess estimation. The optimal locations for thedeployed sensors are uniformly distributed in thefield. In addition, we propose a suboptimal sensordeployment to estimate the Gaussian random fieldwhich is described as an Ornstein-Uhlenbeck process.We show that the suboptimal deployment strategyfor Gaussian random field is also uniformlydistributed. Several simulations show theperformance of our proposed deployment strategies.

WA4.5 P0513 1200 - 1220Real Time Implementation of ModifiedRepetitive Control Strategy in a DC MotorVijayakarthick MUTHUKUMARAN, SathishbabuSANTHANAM, P K BHABA

Annamalai UniversityIn this paper the Modified Repetitive Control

Strategy (MRCS) is designed and implemented in aDC motor. The MRCS incorporates the idea ofrepetitive control strategy (RCS) which accomplishesperfect asymptotic set point tracking in this process,provided that the period length used in the controlformulation matches the actual period of thereference/disturbance signal exactly. The DC motorsystem is approximated into a First Order Plus TimeDelay (FOPTD) model by step testing method. RCS isincorporated in the DC motor control loop ofproportional (P) mode. The proportional controllerparameter is obtained using Ziegler-Nichols TuningRule (ZNTR) .The proposed MRCS is also integratedto a DC motor system. A periodic input signal of sinewave is generated and real time runs of the DCmotor system are carried out for the periodicreference tracking with MRCS based P mode controlloop. A similar run is also carried out with both RCSbased P mode and conventional P-mode controlstructure in the loop. The experimental results showthat the MRCS demonstrated good trackingperformance. A robustness of the MRCS is alsovalidated.

27

Session WA5Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Nightingale

Man-machine InteractionsChairs : Juergen ROSSMANN

RWTH Aachen UniversityWenjie CHENSingapore Institute of ManufacturingTechnology

WA5.1 P0656 1040 - 1100Intuitive Robot Tool Path Teaching using Laserand Camera in Augmented RealityEnvironmentChuen Leong NG, Teck Chew NG, Thi Anh NgocNGUYEN, Guilin YANG, Wenjie CHENSingapore Institute of Manufacturing Technology

This paper presents a new intuitive method forrobot tool path teaching in Augmented Reality (AR)environment. Conventional industrial robot teachingmethod is long known to be either tedious or requirea highly accurate virtual representation of robot workcell. Our method targets to provide the user with afast and easy way of programming an industrial robotfor useful tasks in a safe environment. In our system,a human robot interaction (HRI) system has beendesigned by fusing information from a camera and alaser ranger finder. The video images provide visualinformation to the user to operate the system,whereas the laser range finder captures the Cartesianinformation of the user intended robot working pathsand trajectories. Furthermore, an AR environmenthas been designed where the virtual tool issuperimposed onto the live video. The user simplyneeds to point and click on the image of theworkpiece to generate the tool path. User can alsoadjust virtual tool orientation and simulate the tooltrajectory in the AR environment, thus simplifying therobot teaching task. The proposed system has beentested for robot laser welding application. It isintuitive as no prior knowledge of robotic control isrequired in order to use our system. Mostimportantly, the system is safe and the user does notneed to be physically close to the robot during pathteaching.

WA5.2 P0744 1100 - 1120An FSM based GUI Test Automation ModelYuan MIAO, Xuebing YANGVictoria University

Graphical User Interfaces (GUIs) constitute a largeproportion of today's software and are becomingmore and more complex. Testing the correctness ofGUIs and their underlying software is paramount forproviding quality software products. Manual testing isextremely slow and unacceptably expensive. Wepresent a new technique which enables the processof generating test cases and testing automation,

based on an innovative model. Given a GUI basedapplication, the set of GUI states and their runninglogic is modeled as a finite state machine (FSM). Theefficiency of the model is formally analyzed andcompared with event flow graph (EFG) model. Theresults show that our model is more efficient instorage.

WA5.3 P0826 1120 - 1140A Motion-based Visual Interface for 3DVisualization and Robotic Control ApplicationsSafiullah HUSSAINI, Wooi-Boon GOHNanyang Technological University

The exploratory visualization of 3D data sets ormultiple axis motion control of a monitoring cameramounted on a six-axis robot often require the user tocontrol the viewing orientation and position of acamera view within a 3D space. This paper describesthe design of an inertial motion sensing system,which uses only accelerometers to provide interactive3D control of both orientation and position in suchsystems. We describe the interaction designstrategies for exploratory visualization of 3D datausing the proposed input device, which is mountedbeneath a portable display. Experiments on the useof the proposed device suggest that with sometraining, the users are able to improve the speed withwhich they can navigate the orientation and positionof a virtual camera to a desired target view in the 3Dvisual space

WA5.4 P0909 1140 - 1200Advanced Virtual Testbeds: Robotics KnowHow for Virtual WorldsJuergen ROSSMANNRWTH Aachen University

In recent years, virtual reality has emerged as akey technology for improving and streamliningdesign, programming, manufacturing and trainingprocesses. Based on experiences in the fields ofspace robotics, industrial manufacturing, multi-physics and virtual prototyping, "virtual testbeds" arecurrently being designed and implemented. Buildingon experiences gained in space robotics applications,the idea of "virtual testbeds" currently conquers newfields of applications in the manufacturing industry,on construction sites and even "in the woods".Interestingly, all fields can be supported with a rathergeneric and common basis so that the differentapplications can be realized very cost-efficiently. Thispaper is an attempt to give an overview overdifferent current applications and setups. Fur-thermore, it will focus on the phases of a product'slifecycle that can already today be efficientlysupported by virtual and aug-mented realitytechnologies. From a system theoretic standpoint,virtual testbeds are currently bridging the gapbetween virtual reality and robotics because on theone hand, robotics know how related to kinematicsand physics based modeling is being used in virtualworlds to make them "behave realistically", on theother hand, those virtual worlds make very efficient

28

representations of "real worlds" in which planningsystems for robotics can try out action alternativesquickly and without danger to the robotic hardware.

WA5.5 P0800 1200 - 1220Service Discovery using Software Agents inSemantic WebNandini S SIDNAL, *Ravi S MALASHETTY,**Sunilkumar S MANVIK.L.E.S.C.E.T, Belgaum*VTU, Belgaum**Reva Institute of Technology and Management

The Web, once solely a repository for text andimages but now is evolving into an informationprovider as well as a tool to compute information.With this the internet computing will have a majorrole to play in the service oriented market places.Web Services are one of the fastest growing areas ofinformation technology in recent years, also being amain motivating factor for internet computations inwhich, one of the services being service discovery.Automation of web service discovery is a must as theWWW has become a huge warehouse for storinginformation. This is possible by using software agentsand becomes efficient by enriching web contentssemantically. The Semantic Web (SW) is a mediumwhere the data can be shared and processed both byautomated tools and humans. The key point is in theautomation and integration of the processes throughmachine readable languages. In order to use andconnect all the information available on the Web, thesoftware agents should be able to understand theinformation. The aim of the semantic Web service isto describe and implement Web services so as tomake them more accessible to automated agents.This paper presents an overview of semantic web,software agents and integration of both to achievebetter QoS.

Session WA6Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Pelican

Modeling and IdentificationChairs : Masoud SHAFIEE

Amirkabir University of Technology,TehranLakmal D SENEVIRATNEKing's College London

WA6.1 P0704 1040 - 1100Identification of the Parameters of RobotManipulators Dynamics about an OperatingPoint using Perturbed DynamicsAzeddien KINSHEEL, *Zahari TAHAUniversity of Malaya*University Malaysia Pahang

In this paper, a new approach to identify theparameters of robot manipulator dynamics ispresented. The approach is based on exciting therobot manipulator with multiple identificationtrajectories that pass through the operating pointwith specific speeds and accelerations. The position,velocity and acceleration of the robot joints alongwith the corresponding torque at the operating pointare recorded. The relative change of the joint motionand the corresponding torque compared to areference operating point trajectory is used toidentify the robot dynamics parameters. By usingperturbed rather than the absolute motion values,several modeled and un-modeled effects can beeliminated from the identification equations. This isvery useful when an accurate model of the robotdynamics at specific operating point(s) of theworkspace is required.

WA6.2 P1012 1100 - 1120Descriptor Modeling and Response Analysis ofTwo Rigid-Flexible Cooperative ArmsEhsan SAMIEI, Masoud SHAFIEEAmirkabir University of Technology, Tehran

In this paper, a descriptor dynamic model for atwo-cooperative rigid-flexible link manipulatorssystem which holds a rigid object in a horizontalworkspace is derived. When a system of multimanipulators grasps an object, they form a closedchain that its equations consist of some algebraicconstraints as well as dynamic equations. Thesealgebraic constraints can cause impulsive behaviors inthe system response and must be considered inmodeling and control. Moreover, some of thesealgebraic constraints are dependent to each otherand must be neglected in order to reach to theminimal form of the model. In this paper, a dynamicmodel for a rigid-flexible link manipulator which isderived in many literatures based on Lagrangeequation and assumed mode method is considered.Then, the Minimal constraints equations of the

29

system are derived. The closed chain motion ofmanipulators and constraints are linearized aroundthe rigid-body motion, and a linearized dynamicdescriptor model of the system which is the absolutedescription of the whole system is obtained. Finally,by using descriptor system theory, the behavior ofthe system is investigated. To evaluate the proposedmethod, a simulation was performed on a two-cooperative flexible-link manipulator system.

WA6.3 P1018 1120 - 1140Topology Prpresention and Analysis of CartonManipulationGuowu WEI, Rui Rui ZHANG, Jian S DAIKing's College London

This paper presents models for description andidentification of various cartons in their discretestates, and proposes a new approach to describe thetransformation of configuration states during cartonmanipulation in a packaging process. The methodmakes use of matrix operations which can be used toidentify and model the steps and changes in cartonmanipulation at different stages of a packagingprocess. This gives an analytical way of presentingand identifying information of a carton and ofmodeling carton packaging manipulation andpresents a new way for carton packaging automation.

WA6.4 P1060 1140 - 1200Finite Element Modelling of Rolling Indentationfor Tissue Adomanlity IdentificationKiattisak SANGPRADIT, Kaspar ALTHOEFER, LakmalD SENEVIRATNEKing's College London

We describe a novel approach for demonstrate ofa wheel-rolling tissue deformation as well as theabnormalities tissue depth evaluation using a rollingfinite element model (RFEM). Since a wheeled probewhich is capable of performing rolling tissueindentation has been proven to be a promising deviceto rapid conduct soft tissue property identification forlocalization and documentation of the abnormalitieswithin the tissue, with the aim of compensating theloss of haptic and tactile feedback experienced duringrobotic-assisted minimally invasive surgery (MIS) [3,4, 5]. To implement such a device requires a goodunderstanding of the dynamics of the wheel-tissuerolling interaction and relationship between the tissueinternal structure and the corresponding tissuereaction force. In this paper we propose the RFEM ofthe dynamic interaction between a wheeled probeand a soft tissue sample using ABAQUS finite elementanalysis software package. The aim of this work is tomore precisely locate abnormalities depth within softtissues using RFEM and aid surgeons better in thedecision of resection during MIS through theunderstanding of dynamics of wheel-tissue rollinginteraction. The soft tissue was modelled as anonlinear hyperelastic material with geometricalnonlinearity and the modelling parameters werecalibrated using experimental data from standardtests. The purposed RFEM consists of simulations of

wheel-tissue rolling indentations on a siliconephantom with varied tissue internal structure and alsorunning on a biological tissue such as a porcinekidney. The results show that the proposed methodcan predicted the wheel-tissue interaction force ofthe rolling indentation with a good agreement resultsand the documentary from empirical equation ofRFEM can identify the simulated tumors depthaccurately.

WA6.5 P0356 1200 - 1220The Narcissistic Robot: Robot Calibration usinga MirrorMatthias RÜTHER, Martin LENZ, Horst BISCHOFGraz University of Technology

We present a novel method for calibration of arobotic manipulator. The robot kinematic chain andits tool are observed by a hand mounted camerathrough a mirror. We show the possibility of enablinghand-eye, hand-tool, and kinematic robot calibrationwithout incorporating accurate external references,except the mirror. Using this particularly simplesetup, hand-eye calibration becomes independent ofthe kinematic chain and parameter observabilityconstraints in kinematic calibration become morerelaxed, which makes pose planning for robotcalibration more convenient.

30

Session WA7Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Oriole

Robot Control IChairs : Chien Chern CHEAH

Nanyang Technological UniversityAdel AKBARIMAJDUniversity of Mohaghegh Ardabili

WA7.1 P0509 1040 - 1100A Learning Approach to Optimize WalkingCycle of a Passivity-based Biped RobotNima FATEHI, *Masoud ASADPOUR, **AdelAKBARIMAJD, ***Laila MAJDIQazvin Islamic Azad University*University of Tehran**University of Mohaghegh Ardabili***Islamic Azad University, South Tehran

A learning mechanism based on Powell'soptimization algorithm is proposed to optimizewalking behavior of a passivity based biped robot. Tothis end, a passivity-based biped robot has beensimulated in MSC ADAMS and a control policy inspiredfrom humanoid walking is adopted for a stablewalking of the robot. Linear controllers try to controlthe joints of robot in each walking phase toimplement the gait proposed by the control policy.Learning is employed using Powell's optimizationalgorithm to adjust the control parameters so thatthe robot enters to an optimum limit cycle in a finitetime. The fitness function is defined to evaluate therobot's optimum behavior. The results are verified bysimulations in SIMULINK+ADAMS.

WA7.2 P0584 1100 - 1120Bipedal Robot Walking Strategy on InclinedSurfaces using Position and Orientation basedInverse Kinematics AlgorithmFariz ALI, Aliza CHE AMRAN, Atsuo KAWAMURAYokohama National University

This paper proposes a strategy for bipedal robotwalking on inclined surfaces using position andorientation based inverse kinematics algorithm. Someresearchers implemented control approaches to solvebipedal walking on inclined surfaces. Generally, mostof them apply control feedback at ankle joints andalso introduced many more control methodologies. Inthis paper, inverse kinematics methodology isintroduced systematically for bipedal walking oninclined floor. Positions and orientations areembedded into the kinematics calculation. In thisstrategy, a working bipedal robot walking pattern forflat floor must be developed first. Then, the samewalking pattern can be used for the inclined floorwith orientation included so that the bipedal robot isable to walk on the inclined floor successfully. Thismethodology will distribute the angles caused by theinclined surfaces to the appropriate robot joints. By

doing this, control at ankle joints only is avoided. A 3-D dynamics simulator which is known as RobotControl Simulator and developed in our laboratory isused for simulation in order to validate our proposedmethod.

WA7.3 P0267 1120 - 1140Experimental Investigation of FundamentalProperties of Snake Robot LocomotionPål LILJEBÄCK, *Kristin Y PETTERSEN, *ØyvindSTAVDAHL, *Jan Tommy GRAVDAHLSINTEF ICT*Norwegian University of Science and Technology

This paper derives and experimentally investigatesfundamental properties of the velocity of a snakerobot conducting lateral undulation. In particular, thederived properties state that the average forwardvelocity of the snake robot 1) is proportional to thesquared amplitude of the sinusoidal motion of eachjoint of the robot, 2) is proportional to the angularfrequency of the sinusoidal motion of each joint, 3) isproportional to a particular function of the constantphase shift between the joints, and 4) is maximizedby the phase shift between the joints that alsomaximizes the particular phase shift function. Thepaper presents an experimental investigation of thevalidity of these derived properties by measuring theforward velocity of a physical snake robot duringlateral undulation. The experimental results supportthe theoretical findings.

WA7.4 P0717 1140 - 1200NARMA-L2 Controller for 2-DoF UnderactuatedPlanar ManipulatorAdel AKBARIMAJD, Solmaz KIAUniversity of Mohaghegh Ardabili

NARMA-L2 controller is designed for a 2R planarunderactuated robot. To this end, in systemidentification stage, after appropriate selection ofsystem output, the network is trained in order toapproximate dynamics of the manipulator. Thencontrol law is determined to make the system outputfollow the reference input. Tracking test anddisturbance tests are performed by doing Simulationsin SimMechanics. The results show that the controllerstabilizes the system.

WA7.5 P0445 1200 - 1220Cell Sorting with Combined Optical Tweezersand Microfluidic Chip TechnologiesXiaolin WANG, Zuankai WANG, Dong SUNCity University of Hong Kong

Sorting of specific target cells is an importantprocess in biotechnological research and clinicalmedicine. This paper proposes a methodology thatintegrates optical tweezers and microfluidic chiptechnologies to realize the automatic cell sorting in acontinuous flow environment. In the proposedsystem, cells are driven through the region ofinterest, and the digital image processing technique

31

is utilized to recognize the target cells. The opticaltweezers are used to move the cells selected byimage processing to the desired area. In order tomove the target cells to the collection reservoir morequickly and reduce the cell-cell interactions (e.g.,clustering or jamming), the motion of opticaltweezers is further investigated. The relationshipbetween the laser power and the cell maximal escapevelocity has been studied to achieve the robust cellsorting with lower power. The idea of multi-trapparallel processing is proposed to achieve highthroughput without losing the purity. Utilizing theproposed cell sorter, we can collect rare cells from asample of primary tissue without expansion, with lessdamage and higher accuracy.

Session WA8Date : Wednesday, 08 December 2010Time : 1040 - 1220Venue : Bluebird

Vision ApplicationsChairs : Eric SUNG

Nanyang Technological UniversityYing SUNNational University of Singapore

WA8.1 P0224 1040 - 1100On Line Quality Inspection in Tailor WeldedBlank based on Laws Texture Energy andStructured LightChengning ZHANG, Min XU, Mingyang ZHAO, HaiboLUOChinese Academy of Sciences

Tailored blank laser welding has become a criticaljoining process in a number of manufacturingindustries, including automotive, medical, andaerospace. Even if the welding process involves veryhigh-tech equipment, the final product is notnecessarily free of defects such as lack of fusion orpenetration and porosities. This is the reason why itis important to inspect the welds quality. Theconventional method for the quality inspection ofweld seams is based on the structured lightdeformation, however, if a weld seam hasn’t a grooveprofile, and this method is limited. A new methodbased on combination of Laws texture energy andstructured light is proposed in this paper. A weldseam region and a structured light region areextracted respectively, and then the intersectionoperation with respect to the two regions can becarried out. The given method can also inspect weldseams without structured light deformation.Experiments are conducted with this method, thenthe considered width measurement is well measuredand weld defects can also be detected.

WA8.2 P0918 1100 - 1120Computed Tomography Guided Laparoscopy:Proof of ConceptAdrian PARK, Erica SUTTON, Raj SHEKHARUniversity of Maryland School of Medicine

Current generation minimally invasive surgeriespresent many visualization challenges, including two-dimensional representation of three-dimensionalanatomy and a lack of visualization of deeplyrecessed structures. Coupled with the loss of tactilefeedback which places greater emphasis on availablevisual cues, improved surgical visualization remains along-standing need. Our response to address thisneed is Live Augmented Reality (Live AR), in whichprocessed images from live radiologic scans of thesurgical field are merged with optical images,accounting for spatial and temporal registration. Wehave demonstrated the feasibility of Live AR, but itsclinical implementation is hampered by many current

32

technical limitations. In this report, we havepresented guidance of a simple laparoscopicmaneuver completely based on computedtomography (CT) scanning and rapid 3D rendering ofthe acquired images in the CT room. The capabilitydeveloped here and the reported results constitute astep toward the eventual goal of routine clinicalimplementation of Live AR.

WA8.3 P0944 1120 - 1140Exposure Balancing and Difference Blurring toEliminate Seam-lines in a Real-time Bird's EyeView MonitorSangseok HONG, JinSoo LEE, *Sang-Bok CHOIPohang University of Science and Technology*HANSERO FA Co., Ltd.

A Bird's Eye View Monitor usually shows visibleseam-lines. Previous seam-line elimination methodslimit the dynamic range of mosaicked image andrequire that the reference images are equalgeometrically the regions that overlap. In this paper,we propose a camera model based alternativeexposure compensation algorithm which controlsbrightness of reference images using calculatedoffset, and a seam-line difference blurring algorithmwhich gradually spreads the pixel by pixel differenceof a seam-line over across the seam-line with nogeometrical assumption. Moreover, we generalize thecompensation algortihm for a circularly positionedand consecutively positioned N reference imagemosaicking system. Our approach aims at extendingthe dynamic range, eliminating ghost images, andincreasing system efficiency. Proposed algorithms runon real-time systems. Results are demonstrated usingseveral images.

WA8.4 P0881 1140 - 1200Correction of Paradoxical Vision by SimulatedDepth Cue and Inverted Mirror Image forLaparoscopic SurgeryTakuro ISHII, Chang-Ping KU, Tatsuo IGARASHI,*Harufumi MAKINO, **Yukio NAYAChiba University*Naruto Hospital**Teikyo University Chiba Medical Center

In laparoscopic surgery, surgeons often encounterparadoxical vision according to their position againstthe camera position. Such a paradoxical vision evokesconfusion and surely deteriorates surgicalperformance. Previous researches indicated invertedmirror image is useful to compensate this problemthough upside-down inversion makes depth sensationperplex. To solve the dilemma, we proposed modifiedmethod that displays inverted mirror image plusperspective projection that adds simulated depth cue.After preparing the image adjustment software,trainer box, touch panel device to measure themotion of the forceps, we tested its validity in 36participants including 10 experienced surgeons. Eachparticipants were requested to push buttons followingcomputer fs assignment for ten times of task atthree conditions: 1. stands parallel to the camera, 2.

stands at the opposite side of the camera, 3. standsat the opposite side of the camera watching invertedmirror image with perspective projection. Meanduration of time for completion of the task was23.1+/-5.0 seconds for the subjects stand parallel tothe camera. It was 98.0+/-96.5 seconds and 43.3+/-24.3 seconds for the subject stands at the oppositeside of the camera without and with transformedimage, respectively. Paradoxical vision significantlydeteriorated performance (p<0.001), however, it wascompensated by the inverted mirror vision withperspective projection significantly (p<0.0015). In

grope passing h and gbead drop h trial,tested by 10 experienced surgeons, inverted andperspective vision significantly compensateddeteriorated performance under paradoxical vision.Inverted mirror image with perspective projectionwould be useful tool for correction of paradoxicalvision in laparoscopic surgery, and further researchwould be warranted to make such system practical.

WA8.5 P0349 1200 - 1220Retinal Blood Vessel Segmentation via GraphCutAna G SALAZAR-GONZALEZ, Yongmin LI, Xiaohui LIUBrunel University

Image analysis is becoming increasingly prominentas a non intrusive diagnosis in modernophthalmology. Blood vessel morphology is animportant indicator for diseases like diabetes,hypertension and retinopathy. This paper presents anautomated and unsupervised method for retinal bloodvessels segmentation using the graph cut technique.The graph is constructed using a rough segmentationfrom a pre-processed image together with spatialpixel connection. The proposed method was testedon two public datasets and compared with othermethods. Experimental results show that this methodoutperforms other unsupervised methods anddemonstrate the competitiveness with supervisedmethods.

33

Session WM1Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Ball Room

Panel Session on “Imaging and Vision:Challenges and Opportunities”Chairs : Professor Kar-Ann TOH

Yonsei University, Korea

Panelists:Professor Zhi-Qiang LIUCity University of Hong Kong, SAR, ChinaProfessor Brian C. LOVELLUniversity of Queensland, AustraliaProfessor Massimo TISTARELLIUniversity of Sassari, ItalyProfessor Zhengyou ZHANGMicrosoft Research (MSR), USA

Session WM2Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Canary I

Process ControlChairs : Vaidehi VIJAY

Anna University, ChennaiPeng YANGHebei University of Technology

WM2.1 P0197 1330 - 1350Relative Gain Array for MIMO ProcessesContaining Integrators and/or DifferentiatorsWuhua HU, Wen-Jian CAI, Gaoxi XIAONanyang Technological University

Limitations are revealed for the existing methodsto derive relative gain array (RGA) for a MIMOprocess containing integrators and/or differentiators.This paper proposes a new method to overcomethese limitations. The method can handle processesdescribed by either transfer functions or state-spacemodels. The effectiveness and simplicity of theproposed method are illustrated with examples wherethe existing methods fail to give valid RGAs.

WM2.2 P0223 1350 - 1410Overriding Control for Stability with ManifestVariablesTri TRANUniversity of Technology, Sydney

This paper presents a new method of stabilizingclosed-loop systems, whereby instability warningsand overriding controls are performed. The systemstability is maintained by having a separated agent,here called stabilizing agent, to produce correctiveactions to the manipulated variables. This agentcomputes the stabilizing thresholds on-the-fly basedon dissipative projections and real-time stabilityconditions. It overrides the controller outputs withstabilizing values once the threshold is violated. Theplant models are not employed in this approach.Instead of that, online information of manifestvariables is utilized to evaluate necessary bounds onmanipulated variables for the stability purpose.Theoretical results are illustrated in simulation of amodel predictive controller for an unit operation ofthe paper mill.

WM2.3 P0552 1410 - 1430Output Feedback Control for Nonlinear Systembased on Local Linear Model Trees (LOLIMOT)Yan ZHANG, Linan ZU, Peng YANG, Guolin XINGHebei University of Technology

A nonlinear adaptive output feedback controlapproach based on LOLIMOT is proposed for a classof nonaffine system in the non-strict feedback formwith unknown nonlinearities in this paper. In order to

34

overcome the disadvantages of output feedbackcontrol, LOLIMOT which can identify the nonlinearsystem accurately is applied. During the controlprocess, backstepping approach is utilized to designthe adaptive output feedback controller and theinternal pseudo-controller is used as one-steppredictor. The effectiveness of proposed scheme isshown by simulation on a proper nonlinear system.

WM2.4 P0850 1430 - 1450Applications of Convex Optimization in Plant-wide Control of Membrane Distillation Bio-Reactor (MDBR) Water Recycling PlantAvinash VIJAY, Keck Voon LING, Anthony GORDONFANENanyang Technological University

The objective is to develop a control system thatwill enable the Membrane Distillation Bio-Reactor(MDBR) water recycling plant to become selfsufficient and fully automatic. In order to ensurecontinuous operation, the control system mustmaintain conditions required for the micro-organismsto survive. These requirements need to be met evenwhen solar radiation and weather conditions are notconducive for water production. Hence, maintenanceof a back up reserve is necessary. A balance has tobe struck between minimization of power consumedand maximization of output produced to facilitatesmooth operation. Control is implemented through ahierarchical framework that keeps track of thedifferent plant objectives. Operational set points aredetermined through convex steady state optimization.Current set points are based on predictions regardingvariation in solar radiation within the planninghorizon.

WM2.5 P0926 1450 - 1510First and Second Moment Control of LinearStochastic Discrete Time SystemsFateme ROHANI, *Hamid KHALOOZADEH, **RoyaAMJADIFARDIslamic Azad University, Tehran*K. N. Toosi University of Technology**Tarbiat Moallem University, Tehran

This paper deals with control of first and secondmoment of the states of linear stochastic discrete-time systems. In this paper, the expectation andcovariance of the states are combined together toconstruct a new higher order system. Our approachtends to use the covariance and the expectation ofthe states as feedback signals to obtain theappropriate control signal which leads the states ofthe open loop system to a prescribed value forexpectation vector and covariance matrix. Theresulted system is deterministic, so we would be ableto choose any well-defined control design method. Inthis paper, the method is based on integral controlapproach. The expectation and covariance of anystate of the system are guaranteed to track thedesired corresponding values.

WM2.6 P1105 1510 - 1530Investigation of Thermal Effect in PermanentMagnet Linear Motor StageJ H CHOW, *Zhaowei ZHONG, Wei LIN, *Li PhengKHOO, Wen-Jong LIN, Guilin YANGSingapore Institute of Manufacturing Technology*Nanyang Technological University

Thermally induced deformation has become moreof a concern for precision machines as accuracy isgetting more stringent. Unlike other indirect driveswhich are possible to isolate or place rotating motors,which are heat sources, away from the working area.A linear motor is a direct drive system which is heatsource and placed near to the working area.Therefore, this thermal effect characteristic study isconducted here to find out the effect of heat that istransferred and dissipated around the working area,causing unwanted deformation. The surface areajoining the linear motor coil to the applicationcarriage is large. It will be inaccurate to assume aconstant temperature or heat flux through the entiresurface. The temperature of the surface is differentfor different location and time. Experiments areconducted to investigate the surface temperature andheat flux profile in different application processes.Finally, the result of this investigation can define theheat source accurately so that a further study coulddevelop a prediction scheme for the deformation ofthe carriage. This finding can be applied to errorbudgeting for machine design stages.

35

Session WM3Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Canary II

Co-operative ControlChairs : Kamesh SUBBARAO

The University of Texas at ArlingtonJoseph J YAMEUniversity Henri Poincare

WM3.1 P0210 1330 - 1350Asymptotic Synchronization and CollisionAvoidance for Multi-agent FlockingZhiyong CHEN, *Hai-tao ZHANG, *Chao ZHAIUniversity of Newcastle*Huazhong University of Science and Technology

A novel individual-based alignmen/repulsionalgorithm is proposed in this paper for a flock ofmultiple agents. With this algorithm, each individualrepels its sufficiently close neighbors and aligns tothe average velocity of its neighbors with moderatedistances. In both mathematical analysis andnumerical simulation, we prove that the algorithmguarantees an uncrowded flocking behavior withasymptotical velocity synchronization whensufficiently intensive communication exists within theagents. Moreover, we provide the conditions forcollision avoidance along the whole transientprocedure. The proposed flocking model has itsreferences in natural collective behaviors likeescaping panic and traffic jam motions.

WM3.2 P0249 1350 - 1410Weight Balance for Directed Networks:Conditions and AlgorithmsYuan FAN, *Gang FENG, Yong WANGUniversity of Science and Technology of China*City University of Hong Kong

Consensus strategies find extensive applications incoordination of robot groups and decision making ofagents. Since balanced graph plays an important rolein the average consensus problem for directedcommunication networks, this work explores theconditions and algorithms for the digraph balancingproblem. It has been proved that a directed graphcan be balanced if and only if the null space of itsincidence matrix contains positive vectors. Then twoweight balance algorithms have been proposed, andthe conditions for obtaining a unique balancedsolution have been investigated. This work has alsopointed out the relationship between the weightbalance problem and the features of thecorresponding underlying Markov chain. Finally, twonumerical examples are presented to verify theproposed algorithms.

WM3.3 P0282 1410 - 1430Robust Cooperative Leader-Follower Formation

Flight ControlWeihua ZHAO, Tiauw Hiong GONanyang Technological University

This paper addresses the application of ModelPredictive Control (MPC) approach for the Leader-followers formation flight problem. Under the robustdecentralized unified MPC framework, the currentcollision avoidance scheme has been extended totake care of any shape and small pop-up obstacles.And in the heterogeneous Leader-follower situation, ifthe formation error term is added to the Leader’s costfunction, then the formation will be maintained. Thesimulation results show that the modifieddecentralized MPC framework can successfullyachieve and keep collision-free formation flights.

WM3.4 P0500 1430 - 1450Stability Analysis of Local SwarmsLiangshun WANG, Huajing FANGHuazhong University of Science and Technology

This paper considers finite sense ability and finitemotion ability into swarm system. We first construct amodel of local swarms with a class of attraction andrepulsion. Then we analyze their aggregation motionon the basis of the model. It is shown that theindividuals of the two swarms will aggregate andeventually enter into a bounded hypeball around theswarm center in finite time. We finally analyze theirmotion in convergent region.

WM3.5 P0529 1450 - 1510Asynchronously Dynamic Averaging Estimationwith Communication ConstraintsDequan LI, Xiaofan WANGShanghai Jiaotong University

The paper concerns with the problem ofasynchronous gossip-based dynamic averagingestimation with communication constraints, whereeach agent estimates the local time-varyingparameters individually, then random pairs ofconnected agents iteratively and locally perform apairwise average of their estimations throughquantized information communication. Howquantization affects the evolution of the gossip-baseddynamic averaging estimation algorithm isinvestigated. We prove that the agents’ statesconverge to a random variable that deviates from theaverage of the estimated parameters. We derive astrong result about the upper bound for theasymptotic mean square error of the states, whichjust captures effect of the quantized precision and isindependent of the network parameters.

WM3.6 P1071 1510 - 1530A Constrained Dynamical Systems Approachfor Attitude Consensus of Multiple Rigid BodiesDivya BHATIA, Kamesh SUBBARAOThe University of Texas at Arlington

This paper presents a constrained dynamicalsystems based attitude synchronization and tracking

36

for multiple rigid bodies. The attitude dynamics ofmultiple vehicles are coupled via explicit constraintfunctions that mirror a priori communicationtopologies. Additional constraint functions areimposed to achieve specific consensus and formationobjectives. The constraint stabilization is achieved viaa modified Baumgarte stabilization method thatincludes an integral term. The performance of theseconstrained dynamics based formations is evaluatedfor bounded control authority.

Session WM4Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Kingfisher

Adaptive ControlChairs : Koshy GEORGE

PES Institute of TechnologyKai ZENGERAalto University

WM4.1 P0422 1330 - 1350Well-Posedness of Two Classes of SingularDistributed Parameter Systems in HilbertSpaceFeng LIU, *Guodong SHI, **Delin CHUJiangsu Teachers University of Technology*Changzhou University**National University of Singapore

One of the most important problems for the studyof singular distributed parameter systems is the well-posedness. Not only is it very important for the studyof stability of singular distributed parameter systems,but also it is the theoretic basis for the study of therelated problem of optimal control. In this paper, theconcepts and the properties of generalized operatorsemigroup(GOS) and generalized integralsemigroup(GIS) are given in Hilbert space, thesolving problem of the non-homogeneous singulardistributed parameter system is discussed by theconcepts and the properties of GOS and GIS inHilbert space,and some important results of the twoclasses of singular distributed parameter systems aregiven.

WM4.2 P0514 1350 - 1410Feedforward Multiple Harmonic Control forPeriodic Disturbance RejectionKai ZENGER, *Ali ALTOWATI, **Kari TAMMI, EeroVESAOJAAalto University*University of Garyounis**VTT Technical Research Centre of Finland

The paper discusses compensation of severalsinusoidal disturbance signals entering the systemsimultaneously. The disturbances may have differentfrequencies, amplitudes and phases. A modificationof the well-known LMS algorithm is proposed, whichmakes a direct feedforward compensation algorithmpossible. It is demonstrated that the problems of thestandard LMS algorithm, when several disrurbancefrequencies enter the system simultaneously, cannow be avoided with a multidimensional LMSalgorithm. The performance is demonstrated byextensive simulations.

WM4.3 P0816 1410 - 1430Adaptive Control of a Class of NonlinearSystems using Support Vector Regression

37

Koshy GEORGE, *Prashanth HARSHANGIPES Institute of Technology*PES Centre for Intelligent Systems

In this paper we demonstrate an improvement inthe transient tracking performance when a SupportVector Regression (SVR) is used to identify anonlinear ARMA plant. We use an on-line version ofSVR, and at each instant, the identified model is usedto determine the appropriate control law. A furtherimprovement in the transient performance is shownwith the methodology of multiple models, switching,and tuning.

WM4.4 P0904 1430 - 1450A Sub-Region Boundary-Based Control Schemewith a Least-Squares Estimation Algorithm foran Underwater Robotic SystemZool H ISMAIL, Matthew W DUNNIGANHeriot-Watt University

A new control concept is proposed for anunderwater vehicle-manipulator system (UVMS)where the desired target is defined as a regionboundary rather than a point or a region. For amapping of the uncertain persistent effects i.e.: therestoring forces, a least-squares estimation algorithmalong with the inverse Jacobian matrix is utilized inthe adaptive control law. The unit quaternionrepresentation is used for the attitude representation.The stability analysis is carried out using a Lyapunov-like function. Simulation results are presented toassess the effectiveness of the proposed controlscheme.

WM4.5 P0905 1450 - 1510An Adaptive Region Boundary-Based ControlScheme for an Autonomous UnderwaterVehicleZool H ISMAIL, Matthew W DUNNIGANHeriot-Watt University

A new control concept is proposed for anunderwater vehicle-manipulator system (UVMS)where the desired target is defined as a regionboundary rather than a point or a region. For amapping of the uncertain persistent effects i.e.: therestoring forces, a least-squares estimation algorithmalong with the inverse Jacobian matrix is utilized inthe adaptive control law. The unit quaternionrepresentation is used for the attitude representation.The stability analysis is carried out using a Lyapunov-like function. Simulation results are presented toassess the effectiveness of the proposed controlscheme.

WM4.6 P0876 1510 - 1530Decentralized Adaptive Practical Tracking ofNonlinear Interconnected Systems withDynamic Input and Output InteractionsLixia HAN, Huijin FANHuazhong University of Science and Technology

In this paper, the adaptive practical output

tracking problem on a class of nonlinearinterconnected systems is considered, in which thedynamic interactions and unmodelled dynamicsdepend on both the subsystem inputs and outputs.Under the assumption that both the references andtheir first-order derivatives are bounded,decentralized adaptive backstepping controllers havebeen proposed to make sure that all the signals inthe interconnected systems are globally uniformlybounded, with the tracking errors converging to asmall set around the origin.

38

Session WM5Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Nightingale

Process AutomationChairs : Manuel DE LA SEN

Euskal Herriko UnibertsitateaManish M PATILMAE College of Engineering, Pune

WM5.1 P0106 1330 - 1350Basics on Stabilization of Discrete SwitchedSystemsManuel DE LA SEN, *Asier IBEASEuskal Herriko Unibertsitatea*Universidad Autónoma de Barcelona

This paper is devoted to discuss the role of theminimum residence time in the stabilization ofdynamic discrete switched systems. It is found that itcan be obviated if the sampling period is sufficientlylarge even without imposing the known requirementin the continuous-time case that the various switchedparameterizations (also often refereed to as"configurations") possess a common Lyapunovfunction. However, it is claimed that the same rolerelated to needing a sufficiently large minimumresidence time at each stable parameterization in thecontinuous-time case is now played by the need of asufficiently large minimum stabilizing samplingperiod. This requirement is removed if the variousparameterizations possess a common Lyapunovfunction.

WM5.2 P0245 1350 - 1410Optimization of Reclaiming Voxels for QualityGrade Target with Reclaimer MinimumMovementTien-Fu LU, Maung Thi Rein MYOThe University of Adelaide

This paper presents the first part of study on theproposed automatic optimal reclaiming system. Theproposed system focuses on how the reclaimerreclaims iron ore from stockpiles to fulfill qualitygrade target aiming for minimum reclaimer overallmovement. For example, iron ore can be excavatedfrom open-cut pits, stacked onto stockpiles andreclaimed later using bucket wheel reclaimer (BWR)for export to customers. Stockpiles here arerepresented by combination of voxels eachpossessing certain quality distribution. The supply ofiron ore closest to the target grade with loweroperation cost is the main concern for the oreproducer. Currently, BWRs are operated manually tomeet quality grade demand involving the checking ofquality periodically during reclaiming operation. Inresult, there is excessive movement of the largemachine leading to higher energy consumption. Inthis paper, the combination of voxels is identifiedfrom available stockpile voxels considering BWR

minimum movement to fulfill quality and quantitydemand of iron ore. Achieving the desired grade withmaximum efficiency will result in increasingproductive rates and less impact on global warmingby significantly reducing energy consumption.

WM5.3 P0737 1410 - 1430Automated Modeling and Analysis of Agent-based Simulations using the CASE FrameworkJames DECRAENE, Malcolm Yoke Hean LOW,Fanchao ZENG, Suiping ZHOU, Wentong CAINanyang Technological University

We present a modular evolutionary framework,coined CASE for “complex adaptive system evolver'',to automate the modeling and analysis of agent-based simulations (ABSs). The field of agent-basedmodeling is rapidly growing due to its capabilities toexpose the emerging complex phenomena occurringin a wide range of natural and artificial systems suchas biological cells, societies, battlefields, stockmarkets, etc. Nevertheless, studying agent-basedsimulations is a complicated, interdisciplinary andtime-consuming process. Indeed, a large number ofsimulation parameters has to be considered toidentify and fully understand the conditions leading tothe emerging phenomena of interest. To tackle thisdifficulty, the study of ABSs is thus typicallyconducted in an iterative manner, where eachiteration includes the successive and manualmodeling of ABSs and analysis of simulationoutcomes. To automate this iterative and time-consuming process, we propose CASE, a platform-independent framework capable of evolving ABSs toexhibit the desired emerging behaviors. Through thisevolutionary approach, the examination, i.e., themodeling, execution and analysis, of ABSs isautomated. This process automation significantlyfacilitates the examination of complex systems usingABSs. In this paper, we present in detail this modularevolutionary framework which is illustrated with anexample experiment. In this experiment, CASE isutilized for Automated Red Teaming, a simulation-based vulnerability assessment technique commonlyemployed by defense analysts. The aim of this paperis to introduce this flexible computational frameworkwhich may potentially benefit related fields involvingagent-based simulations such as the gaming orfinancial industries.

WM5.4 P0902 1430 - 1450A FDD Method by Combining Transfer Entropyand Signed Digraph and its Application to AirSeparation UnitQian HOU, Lei WANG, Ningyun LU, Bin JIANG,*Jianhua LUNanjing University of Aeronautics and Astronautics*Southeast University

A fault detection and diagnosis (FDD) method isproposed by combining transfer entropy (TE) andsigned digraph (SDG). Given process historical data,transfer entropy is used to construct a SDG model torepresent causal relationship between process

39

variables. A fault severity evaluation method is thenproposed based on the modified SDG model, wherethe nodes can take values (±1), (±3) and (±6).Then, an index named DoF is developed to measurefault severity. The application results can verify theeffectiveness and feasibility of the proposed method.

WM5.5 P0484 1450 - 1510A Multi-Purpose Remote Controller based onBluetooth Mobile PhoneChih-Hung WU, Li-Shan MA, Yu-Jyun CAI, Ting-FuYEHChienkuo Technology University

In this study, a multi-purpose remote controllerbased on Bluetooth mobile phone (MRCBMP) isproposed to provide users for a convenient controlinterface. The IR, RF and Bluetooth modules areintegrated into the microcontroller modulecooperated with Bluetooth mobile phone by Bluetoothmodule. By MRCBMP, users can control IR, RF andBluetooth controlled objects without the numerousconventional remote controllers.

WM5.6 P1026 1510 - 1530Speed Optimization for Micropipette Motionduring Zebrafish Embryo MicroinjectionShengfeng ZHOU, Peter C Y CHEN, *Zhe LU, Joo HooNAM, **Hong LUO, Ruowen GE, Chong Jin ONG,**Wei LINNational University of Singapore*University of Toronto**Singapore Institute of Manufacturing Technology

This paper presents an approach to optimize thezebrafish-embryo microinjection process by selectingsuitable speed trajectories for the motion of themicropipette, with the objective of minimizing the riskof damaging the embryo during the microinjectionprocess. A viscoelastic model of a zebrafish embryo isdeveloped based on experimental data. The speedoptimization problem is formulated based on thisviscoelastic model. Simulation results indicate thatthere exists an optimal speed trajectory for themicropipette under certain conditions. The keybenefit of this speed optimization approach is thatthe turgor pressure inside the membrane isprevented from increasing significantly during themicroinjection process, thus avoiding damaging thebiological components in the embryo.

Session WM6Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Pelican

Mobile Robotics IChairs : Maulin M JOSHI

Sarvajanik College of Engineering andTechnology, Surat

WM6.1 P0227 1330 - 1350Experiences with Online Local Model PredictiveControl for WMR NavigationLluís PACHECO, Ningsu LUOUniversity of Girona

Path-tracking by using MPC (model predictivecontrol) is a suitable control science solution formobile robot navigation applications. Online MPC isreported by using short-term horizons that allowdealing with flexible path-tracking and reactivebehaviors. The majority of MPC experimentalresearch developed is based on the fact that thereference trajectory is known beforehand. However,under dynamic environments the global solutionbecomes unfeasible for the majority of applicationswhere the scenario should be considered as partiallyunknown due to the lack of global sensors or theexistence of dynamic obstacles. Moreover, traditionalmotion control of wheeled mobile robots (WMRs) isachieved by using discontinuous control laws that areimplemented through low level velocity PIDcontrollers. Instead of using such methods, this workproposes to use local MPC as a useful methodologyfor WMR navigation under dynamic environments oras obstacle avoidance strategy. In this way, thedesirable path coordinates are used in the control lawas a way for obtaining the robot speed commands.Therefore, simulation results are used for addressingonline MPC implementations. The system is on-robottested by using simple on board perception systems.In this context, a local occupancy grid is built byusing dead-reckoning and monocular data.

WM6.2 P0357 1350 - 1410A Robust Downward-Looking Camera BasedVelocity Estimation with Height Compensationfor Mobile RobotsXiaojing SONG, Kaspar ALTHOEFER, Lakmal DSENEVIRATNEKing's College London

Slip plays a vital role in traction control when amobile robot traverses over soft soils. To estimateslip parameters, accurate measurement of robotvelocity is particularly required. Previous related workdone by the authors has adopted a single downward-looking single camera system for velocity and slipestimation [1][2]; however, such a single camerasystem is prone to lose accuracy when the distancebetween the camera and terrain is time-varying, suchas traversing over uneven terrains [1]. To cope with

40

the problem, this paper presents a robust downward-looking camera based velocity estimation approach,which can particularly be capable of identifying heightvariation and compensating for velocity estimation. Adownward-looking stereo camera instead ofpreviously used single camera is adopted. Thecamera-terrain distance can be estimated bymatching same features in left and right frames.Robot velocity measured with height compensationcan be more accurate, compared to estimateswithout it. The proposed approach has beenvalidated through comprehensive experimental studyon a lab-based test rig; and test results show goodperformance of the proposed approach. With theproposed method, slip estimation techniques given by[2][3] can be promisingly extended to non-flatterrains.

WM6.3 P0553 1410 - 1430Neuro-Fuzzy Based Autonomous Mobile RobotNavigation SystemMaulin M JOSHI, *Mukesh A ZAVERISarvajanik College of Engineering and Technology,Surat*Sardar Vallabhbhai National Institute of Technology,Surat

Neuro-fuzzy systems have been used in past yearsfor robot navigation applications because of theirability to learn human expertise and to utilize thisknowledge to develop autonomous navigationstrategies. In this paper, neuro-fuzzy based systemsare developed for behavior based control of a mobilerobot for reactive navigation. The proposed systemstransform sensors’ input to yield wheel velocities.Novel algorithms are proposed for a) to find therange of the mobile robots from nearby obstacles andb) to generate training pairs for neural network,optimally. With a view to ascertain the efficacy ofproposed system; developed neurofuzzy system’sperformance is compared to neural and fuzzy basedapproaches. Simulation results show effectiveness ofproposed system in all kind of obstacle environments.

WM6.4 P0795 1430 - 1450An Effective Collaborating Algorithm forSwarm Robots Communicating bySematectonic StigmergyTuan A PHAN, R. Andrew RUSSELLMonash University, Melbourne

This paper investigates an algorithm, whichimproves the task completion rate for simple swarmrobots implementing a leaf-curling task. In thisbiologically inspired task, robots collaborate to find asuitable place to bend a leaf, which allows them tosuccessfully fold it up. To complete the task simplerobots were developed that are not equipped withany direct communication devices. Theycommunicate via sematectonic stigmergy, whichmeans every robot can only gain information viachanges in their working environment, which aremade by other robots. This type of communicationhas proved beneficial in helping swarm robots

monitor the performance of other swarm memberswithout direct contact, team mate localization orrecognition. However, in earlier experiments,implementing the leaf-curling task, informationperceived by every robot has not been effectivelyused to create meaningful collaboration. Thisdisadvantage becomes evident via the low taskcompletion rate. If robots explore their environment,this will improve the outcome by increasing theprobability of finding the most suitable part of theleaf to work on. In this paper, an algorithm enablingswarm robots to effectively explore the environmentand find the most effective place to perform the leaf-curling task is described in detail. The improvementof completion rate, achieved by this exploring rule, isverified by both simulation and physical experimentswith a group of W-AntBots.

WM6.5 P0908 1450 - 1510X-band Radar Based SLAM in Singapore's Off-shore EnvironmentJohn MULLANE, *Samuel KELLER, Akshay RAO,Martin David ADAMS, **Anthony YEO, ***Franz S.HOVER, ***Nicholas M PATRIKALAKISNanyang Technological University*Swiss Federal Institute of Technology**National University of Singapore***Massachusetts Institute of Technology

This paper presents a simultaneous localisationand mapping (SLAM) algorithm implemented on anautonomous sea kayak with a commercial off-the-shelf X-band marine radar mounted. TheAutonomous Surface Craft (ASC) was driven in anoff-shore test site in Singapore's southern Selat Puahmarine environment. Data from the radar, GPS andan inexpensive single-axis gyro data were logged byan on-board processing unit as the ASC traversed theenvironment, which comprised geographical andsurface vessel landmarks. An automated featureextraction routine is presented, based on aprobabilistic landmark detector, followed by aclustering and centroid approximation approach. Withrestrictive feature modeling, and a lack of vehiclecontrol input information, it is demonstrated that viathe novel RB-PHD-SLAM Filter, useful results can beobtained, despite an actively rolling and pitching ASCon the sea surface. In addition, the merits ofinvestigating ASC SLAM are demonstrated,particularly with respect to the map estimation,obstacle avoidance and target tracking problems.Despite the presence of GPS and gyro data, headinginformation on such small ASC's is greatlycompromised which induces large sensing error,further accentuate by the large range of the radarsensor. This work is a step towards realising an ASCcapable of performing environmental or securitysurveillance and reporting a real-time activeawareness of the above-water scene.

WM6.6 P0984 1510 - 1530On Leader-Based Shape CoordinationReza HAGHIGHI, Chien Chern CHEAH

41

Nanyang Technological UniversityIn this paper, we consider leader-based shape

coordination for multi-robot systems. It is shown thathaving knowledge about desired velocity andeffective interaction among individuals can helpgroup members to continue the path even afterleader failure. Since any collective behavior emergesfrom interaction among individuals, we proposeinteractive force to maintain minimum distanceamong agents as well as group unity duringmovement even after leader failure. Simulationresults are presented to illustrate the performance ofproposed method.

Session WM7Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Oriole

Localization, Navigation and Mapping IChairs : Keck Voon LING

Nanyang Technological University

WM7.1 P0320 1330 - 1350SLAM with Salient Line Feature Extraction inIndoor EnvironmentsSu-Yong AN, Jeong-Gwan KANG, Lae-Kyoung LEE,Se-Young OHPohang University of Science and Technology

This paper presents a simultaneous localizationand mapping (SLAM) of a large indoor environmentusing Rao- Blackwellized particle filter (RBPF) alongwith line segments as the landmarks. To representthe environment as a compact form, we use only twoend points of the line segment, reducingcomputational cost in modeling line uncertainty. Witha modified scan point clustering method, theproposed adaptive iterative end point fitting (IEPF)plays an important role in estimating line parametersby taking a noisy scan point near end points intoaccount. Thus, by line-segment matching the robot islocalized well in a local frame. We also introduce anonline global optimization of a map, which providesmore consistent map by removing spurious lines andmerging collinear lines. Each of our approaches isefficiently integrated into the proposed RBPFSLAMframework. Experiments with well-known data setdemonstrate that the proposed method provides areliable SLAM performance along with a compact maprepresentation.

WM7.2 P0475 1350 - 1410Role-Based Path Planning and Task Allocationwith Exploration Tradeoff for UAVsChristoph RASCHE, Claudius STERN, LisaKLEINJOHANN, Bernd KLEINJOHANNUniversity of Paderborn

After large accidents or natural disasters a fast andefficient retrieval of injured is required. With recentadvances, unmanned aerial vehicles (UAVs) are moreand more a viable choice under such circumstances.The problem of coordination arises with the numberof employed UAVs as well as proper task allocationamong possibly heterogeneous UAVs. In this work ahybrid approach for UAV coordination and taskallocation is presented. It covers the exploration ofunknown terrains as well as role-based, goal-orientedpath planning and simultaneous task allocation. Theapproach uses a role system to support prioritizedtask assignment and an informed search algorithm tofind consistent trajectories to known goal positions.The system is suited for highly dynamicenvironments, requiring frequent recalculations.

42

WM7.3 P0785 1410 - 1430Line-based SLAM with Slow Rotating RangeSensors: Results and EvaluationsDamien VIVET, Paul CHECCHIN, Roland CHAPUISBlaise Pascal University

This paper is concerned with the SimultaneousLocalization And Mapping (SLAM) application with amobile robot moving in a structured environmentusing data obtained from rotating sensors such asradars or lasers. A line-based EKF-SLAM (EKF standsfor Extended Kalman Filter) algorithm is presented,which is able to deal with data that cannot beconsidered instantaneous when compared with thedynamics of the vehicle. When the sensor motion isfast relative to the measurement time, scans becomelocally distorted. A mapping solution is presented,that includes sensor motion in the observation modelby taking into account the dynamics of the system.Experimental results with real-world 2D-laser scannerdata are presented. Moreover a performanceevaluation of the results is carried out. A quantitativeperformance evaluation method is proposed whendealing with a 2D line map and when a ground truthis available. It is based on the bipartite graphmatching and combines several criteria that aredescribed. A comparative study is made between theoutput data of the proposed method and the dataprocessed without taking into account distortionphenomena.

WM7.4 P0883 1430 - 1450Large-Scale Monocular SLAM by Local BundleAdjustment and Map JoiningLiang ZHAO, Shoudong HUANG, *Lei YAN, JackJianguo WANG, Gibson HU, Gamini DISSANAYAKEUniversity of Technology, Sydney*Peking University

This paper first demonstrates an interestingproperty of bundle adjustment (BA), "scale driftcorrection". Here "scale drift correction" means thatBA can converge to the correct solution (up to ascale) even if the initial values of the camera posetranslations and point feature positions are calculatedusing very different scale factors. This propertytogether with other properties of BA makes it thebest approach for monocular SimultaneousLocalization and Mapping (SLAM), withoutconsidering the computational complexity. Thisnaturally leads to the idea of using local BA and mapjoining to solve large-scale monocular SLAM problem,which is proposed in this paper. The local maps arebuilt through Scale-Invariant Feature Transform(SIFT) for feature detection and matching, randomsample consensus (RANSAC) paradigm at differentlevels for robust outlier removal, and BA foroptimization. To reduce the computational cost of thelarge-scale map building, the features in each localmap are judiciously selected and then the local mapsare combined using a recently developed 3D mapjoining algorithm. The proposed large-scalemonocular SLAM algorithm is evaluated using apublicly available dataset with centimeter-levelground truth.

WM7.5 P0889 1450 - 1510Static and Dynamic Fusion for OutdoorVehicule LocalizationBastien VINCKE, Alain LAMBERT, *DominiqueGRUYER, Abdelhafid ELOUARDI, **EmmanuelSEIGNEZUniversité Paris sud*INRETS/LCPC**ESIEE

The vehicle's localization is classically achieved byBayesian methods like Extended Kalman Filtering.Such a method provides an estimated position withits associated uncertainty. Bounded-error approachesusing interval analysis work in a different way as theyprovide a possible set of positions. An advantage ofsuch approaches is that the results are guaranteedand are not probabilistically defined. This paperfocuses on constraints propagation techniques usingstatic and dynamic fusion. Static fusion uses dataredundancy to enhanced proprioceptive data. Thendynamic fusion uses GPS in order to reduce the sizeof the localization box. The approach has beenvalidated with a real outdoor vehicle.

WM7.6 P0615 1510 - 1530Optimal Transmission Power Selection underEnergy Constraints for Sensor NetworkLocalizationP.T.V. Bhuvaneswari MOHAN, *V VaidehiVIJAYAKUMAR, M Agnes Saranya MOHANAnna University, Chennai*Madras Institute of Technology, Chennai

In a wireless sensor network nodes can bedeployed randomly to collect data from an area ofinterest. So it is mandatory for each node in thenetwork to be aware of its location in the physicalworld, in order to map the collected information tothe base station. Thus, localization becomes animportant technique in sensor networks. Further,power conservation is also an important factor as thenodes are operated with limited battery supply.Hence to increase the lifetime of the network, anenergy-efficient localization is required. In this paper,a fine-grained paradigm of localizing sensor nodesbased on Received Signal Strength (RSS) undercontrolled power of anchor nodes is proposed. Thetransmission power control is achieved by DistanceControlled Transmission Power (DCTP) adjustmentalgorithm where an anchor node closer to anunknown node operates at low transmission powerwhile the one at farther distance operate at hightransmission power. Thus based on the distance ofthe unknown node with anchor node, thetransmission power level of anchor is adjusted. Thispaper also provides analysis of the impact of varioustransmission power levels on the proposedlocalization algorithm. Simulation analysis is carriedout in Matlab and the results show the performanceof the algorithm in terms of localization accuracy andpower consumption.

43

Session WM8Date : Wednesday, 08 December 2010Time : 1330 - 1530Venue : Bluebird

Human Centered SystemsChairs : Jean-Yves FOURQUET

ENIT

WM8.1 P0691 1330 - 1350A Haptic Training Environment for the HeartMyoblast Cell Injection ProcedureVu Thanh LE, Saeid NAHAVANDI, Doug CREIGHTONDeakin University

The heart muscle of a cardiac arrest victimcontinues to accumulate damage throughout itslifetime. This reduces the heart’s ability to pumpsufficient oxygen and nutrient blood to meet thebody’s needs. Medical researchers have shown thatdirect injection of pre-harvested skeletal myoblastcells into the heart can restore some muscle function[1]. This operative procedure usually necessitates thesurgeon to open a patient’s chest. The open chestprocedure is usually a lengthy process and oftenextends the recovery time of the patient.Alternatively, a high accuracy surgical aid roboticsystem can be used to assist the thoracoscopicsurgery [2][3]. While the robotic surgical method aidsfaster patient recovery, a less experienced surgeoncan potentially cause damage to surrounding tissue.This paper presents a study into the development ofa virtual haptically-enabled heart myoblast injectionsimulation environment, which can be used to trainnew surgeons to get hands on experience with theprocess. The paper also discusses the development ofa generic constraint motion technique for needleinsertion. Experiments on human performancemeasures and efficacy, while interacting with hapticfeedback training models, are also presented. Theexperiment involved 10 operators, with each personrepeating the needle insertion and injection 10 times.A notable improvement in the task execution timewith the number of repetitions was observed.Operators improved their time by up to 300%compared to their first training attempt for a staticheart scenario. Under a dynamic heart motion,operator’s performance was slightly lower, with thesuccessful rate of completing the experiment reducedfrom 84% to 75%.

WM8.2 P0539 1350 - 1410Hierarchical Learning Approach for One-shotAction Imitation in Humanoid RobotsYan WU, Yiannis DEMIRISImperial College, London

We consider the issue of segmenting an action inthe learning phase into a logical set of smallerprimitives in order to construct a generative modelfor imitation learning using a hierarchical approach.Our proposed framework, addressing the “how-to”

question in imitation, is based on a one-shot imitationlearning algorithm. It incorporates segmentation of ademonstrated template into a series of subactionsand takes a hierarchical approach to generate thetask action by using a finite state machine in agenerative way. Two sets of experiments have beenconducted to evaluate the performance of theframework, both statistically and in practice, throughplaying a tic-tac-toe game. The experimentsdemonstrate that the proposed framework caneffectively improve the performance of the one-shotlearning algorithm and reduce the size of primitivespace, without compromising the learning quality.

WM8.3 P0661 1410 - 1430Saliency Ranking for Benthic Survey usingUnderwater ImagesMatthew JOHNSON-ROBERSON, *Oscar PIZARRO,*Stefan WILLIAMSKTH Royal Institute of Technology*The University of Sydney

This paper presents a novel architecture for aclassification system based on the visual saliency ofimages. The work is motivated by the difficulty ofreviewing large numbers of images as a humanoperator in the context of Autonomous UnderwaterVehicle (AUV) surveys. We formulate a feature spacein which an algorithm operates over color and textureto determine saliency and illustrate how this can beused to find interesting or unusual images within alarge data set. The saliency classification based onthese general image features allows for overlayshighlighting interesting benthos or geologic structureson large scale 3D seafloor reconstructions, quicklyproviding spatial context to human observers. Theseresults are validated using a set of human trials inwhich images are classified into salient and non-salient categories by a number of test subjects. Thetrials show good agreement both between subjectsand between the human labels and the automatedclassification system. The results of the automatedtechnique are also compared directly to a moretraditional SVM classification system showingfavorable results for our system for generalizing tonew environments.

WM8.4 P0965 1430 - 1450On the Collaboration of an Automatic Path-planner and a Human User for Path-Finding inVirtual Industrial ScenesNicolas LADEVEZE, Jean-Yves FOURQUETENIT

This paper describes a global interactiveframework enabling an automatic path-planner and auser to collaborate for finding a path in clutteredvirtual environments. First, a collaborativearchitecture including the user and the planner isdescribed. Then, for real time purpose, a motionplanner divided into different steps is presented.First, a preliminary workspace discretization is donewithout time limitations at the beginning of thesimulation. Then, using these pre-computed data, a

44

second algorithm finds a collision free path in realtime. Once the path is found, an haptic artificialguidance on the path is provided to the user. Theuser can then influence the planner by not followingthe path and automatically order a new pathresearch. The performances are measured on testsbased on assembly simulation in CAD scenes.

WM8.5 P0969 1450 - 1510Automatic Gait Characterization for a MobilityAssistance SystemCong ZONG, Mohamed CHETOUANI, *Adriana TAPUSUniversité Pierre et Marie Curie*Ecole Nationale Supérieure de Techniques Avancées

This paper addresses gait analysis for a mobilityassistance robot designed for the elderly people. Sixpatients and ten healthy peoples were invited to bepart of our first pilot experiment. We designed twoexperiments so as to firstly detect gait parametersand secondly to identify a change of speed. For thefirst trial, we compared the temporal-distanceparameters of the healthy people and of individualssuffering of mobility problems. The percentages ofthe gait cycle for duration of stance are higher forpeople with mobility impairment than for healthypeople. In the second experiment, we detected thechange in walking speed from the ten healthypeoples. Two different metrics derived from theKullback-Leibler (KL) divergence and from theGeneralized Likelihood Ratio (GLR) were employedfor walking change detection. The Receiver OperatingCharacteristic (ROC) curves show a betterperformance for the signal obtained with theaccelerometer sensor than that obtained with theinfrared distance sensor. Nevertheless, the results ofour experiments demonstrated that bothmethodologies (KL and GLR) can be used to detectthe change points during walking at high or slowspeed.

WM8.6 P1065 1510 - 1530Just-in-time Cooperative SimultaneousLocalization and MappingAdrian MARTIN, M. Reza EMAMIUniversity of Toronto

A new technique for Simultaneous Localization andMapping (SLAM) is introduced. This technique wasdeveloped as a real-time, distributed, scalableimplementation for heterogeneous mobile robotteams. Update efficiency and performance undervariable resources is enhanced using a new strategycalled Lazy Belief Propagation. The formulations andalgorithms behind the implementation are describedand a simulation was used to compare several SLAMalgorithms. Results demonstrate an 11%improvement in map coverage in the same amount oftime compared to a traditional implementation.

Session WP2Date : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Canary I

Identification and Estimation IChairs : Kurt E HÄGGBLOM

Åbo Akademi UniversityZhiping LINNanyang Technological University

WP2.1 P0616 1620 - 1640An Unscented Kalman Filtering Approach forNonlinear Singular SystemsShuwen PAN, Hongye SU, Zhitao LIU, *Pu LIZhejiang University*The Technical University of Ilmenau

In this study, an unscented Kalman filteringapproach is proposed for nonlinear singular systemsto obtain not only the estimation for the states butalso for the unknown inputs presented in themeasurement equations. No prior information isneeded for the unknown inputs to be estimated. Theformulation of the proposed approach is based on theweighted least squares estimation (LSE) and theunscented transformation (UT) methods. Therestriction of the proposed approach is alsomentioned. An illustrative example demonstrates thataccurate and consistent state and unknown inputestimations are obtained with the proposedapproach.

WP2.2 P0760 1640 - 1700Identification of Wiener Systems based onFixed Point TheoryGuoqi LI, Changyun WENNanyang Technological University

In this paper, we propose a new method for theidentification of Wiener systems based on fixed pointtheory. The linear part of the system is an infiniteimpulse response (IIR) system and the nonlinearstatic function is allowed to be non-continuous ornon-smooth. Our proposed technique transforms theestimation of parameters to finding a fixed point of anonlinear equation. We show the existence of thefixed point and also develop an iterative algorithm tofind the fixed point. It is proved that, the determinedfixed point is actually a global minimum point of thecost function and it is unique, and thus globalconvergence of the estimates is ensured. Theperformance of the proposed approach is illustratedby simulation studies.

WP2.3 P0867 1700 - 1720Robust Tracking in Mixed LOS/NLOSEnvironmentsLili YI, Chin-heng LIM, *Chong-Meng SEE, SirajudeenGULAM RAZUL, Zhiping LINNanyang Technological University

45

*DSO National LaboratoriesNon-line-of-sight (NLOS) error is one of the most

important factors affecting the accuracy ofpositioning or tracking especially in urban or indoorenvironments. This paper concentrates on alleviatingthe influence of the NLOS error. A position detectionapproach utilizing the circle error probability (CEP) inconjunction with the least square (LS) method andKalman Filter (KF) tracking algorithms is proposed.The LS is used to estimate the positions of a movingtarget from time-ofarrival (TOA) measurements andthe KF is used to smooth the tracking trajectory.Simulation results show that this approach caneffectively identify line-of-sight (LOS) estimatedpositions, leading to higher tracking accuracy thanthat by other tracking methods without using positiondetection.

WP2.4 P1035 1720 - 1740Convex Formulations for Data-BasedUncertainty Minimization of Linear UncertaintyModelsKurt E HÄGGBLOMÅbo Akademi University

Convex formulations are derived for theminimization of uncertainty bounds with respect to anominal model and given input-output data forgeneral uncertainty models of LFT type. The knowndata give rise to data-matching conditions that haveto be satisfied. It is shown how these conditions,which originally are in the form of BMIs for a numberof uncertainty models, can be transformed to LMIs,thus making the optimization problem convex. Theseformulations make it easy to find the best uncertaintymodel from a number of alternatives for robustcontrol design.

WP2.5 P0321 1740 - 1800Noise Covariance Estimation using DualEstimation for Disturbance Storm Time IndexApplicationBoonsri KAEWKHAM-AI, Kasemsak UTHAICHANAChiang Mai University

The disturbance storm time (Dst) index is used forpredicting the geomagnetic storm that can affectmany systems on earth. The application of the dualunscented Kalman filter (DUKF) to improve thequality of the Dst index prediction by simultaneouslyestimating the process noise covariance is set forth inthis paper. In DUKF, two unscented Kalman filters(UKFs) are run in parallel. The UKF applied to amodel-based Dst index prediction is so called a stateestimator; while the other, a parameter estimator, isfor identifying and recursively updating the processnoise covariance. The performance comparisonbetween the traditional UKF with fixed constantvalues of the process noise covariance, and the DUKFare examined. The actual all Dst and the Dst dataduring the storm (below -80 nT) are used to assessthe quality of the predictions. It is found that rootmean square error (RMSE) of Dst index predictionusing the DUKF is lower than that of the UKF with

fixed constant process noise covariances. Specifically,RMSEs of the DUKF are 6.5816 for all Dst and18.0615 for Dst below -80 nT, whereas, theprediction using a fixed constant process noisecovariance yield RMSEs of at least 6.6678 and19.3954 for all Dst and Dst below -80 nT,respectively. Hence, the DUKF outperforms thetraditional UKF with fixed constant process noisecovariances in this study.

46

Session WP3Date : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Canary II

Complex SystemsChairs : Yan-Ping LI

Shenyang UniversitySomanath MAJHIIndian Institute of Technology,Guwahati

WP3.1 P0381 1620 - 1640Stability of Complex Systems with MixedConnection Configurations under SharedControlTri TRAN, Hung T NGUYEN, Q P HAUniversity of Technology, Sydney

This paper presents a new stabilizing method forthe control of complex systems operating in semi-automatic modes. The complex system is modeled byseveral spatially-coupled subsystems interconnectedin parallel, serial and cycle configurations. Eachsubsystem is regulated by a dedicated autonomouscontroller that also allows for a manual control mode.An interconnection stability condition which takes thecouplings between subsystems into consideration isderived from the renowned dissipative systemstheory. Built upon this stability condition,decentralized stabilizing agents for autonomouscontrollers are subsequently deployed independentlyand segregatedly from the control algorithms. Due tothis independence, human errors from manmachineinteractions, that may destabilize the control systems,can be avoidable; also different types of controlalgorithms and controllers of subsystems areinteroperable with the same stabilizing mechanism.To accomplish such tasks simultaneously, thestabilizing agents render overriding outputs for theautomatic controllers, and at the same time, provideinstability warning signals and manipulation guidanceto the operators to successfully regulate thesubsystems in the manual control mode, yet maintainthe plant-wide stability. Real-time data of controlinputs and plant outputs is exerted under theauspices of controller dissipativity indices andtrajectories to stabilize the systems with closed-loopcontrol and man-in-the-loop coexistence. Our mainresults are illustrated in simulation for a three-unitsystem.

WP3.2 P0156 1640 - 1700Controllability of Fractional Impulsive NeutralFunctional Integrodifferential Inclusions withInfinite Delay in Banach SpacesZhixin TAI, *Xingcheng WANGBohai University*Dalian Maritime University

In this work the controllability of fractionalimpulsive neutral functional integrodifferential

inclusions in a Banach space has been addressed.Sufficient conditions for the controllability areestablished using fractional calculus, a semigroup ofoperators and fixed point theorem of Leray-Schaudernonlinear alternative.

WP3.3 P0600 1700 - 1720Routing Strategy with Waiting Time on Scale-free NetworkDan WANG, Yan-Ping LI, Wenwei LIU, *Liang DONG,*Lin LIShenyang University*Northeastern University

We study traffic dynamics and propose a newrouting strategy based on the local topologicalinformation in scale-free network. According to thenew routing strategy with a single tunable parameter,a packet is delivered to next node with probabilitiesthat depend not only on the degrees of the next nodebut also the estimated waiting time at the new node.The routing algorithm is implemented in BA scale-freenetworks in the case of nodes’ delivering abilityproportional to its degree. Simulations show that themaximal network capacity corresponds to an optimalcontrolled parameter.

WP3.4 P0966 1720 - 1740Static Output Feedback Control for ContinuousDescriptor SystemsYu ZHENG, Jun-e FENG, Lin SUN, Juan YAOShandong University

The static output feedback control problem isinvestigated in this paper for continuous descriptorsystems. New necessary and sufficient conditions areestablished to ensure the admissibility of thecontinuous descriptor systems. Then, based on thenew necessary and sufficient conditions, sufficientconditions are derived for admissibility of the closed-loop systems. The desired static output feedbackcontrollers can be derived directly via the conecomplementarity linearization techniques. Finally, anillustrative example is presented to demonstrate theeffectiveness of the proposed approach in this paper.

WP3.5 P0801 1740 - 1800Active Sway Control of a Single PendulumGantry Crane System using Output-DelayedFeedback Control TechniqueRajeeb DEY, Nishant SINHA, Priyanka CHAUBEY,*Sandip GHOSH, **Goshaidas RAYSikkim Manipal University*National Institute of Technology Rourkela**Indian Institute of Technology, Kharagpur

This paper investigates the implementation ofoutput-delayed feedback control (ODFC) techniquefor controlling the sway angle of single pendulumgantry crane (SPGC) system. Linearized mathematicalmodel of the SPGC in state space form is consideredfor the investigation. The designed ODFC hasundergone complete stability analysis for a givencontroller gain.

47

Session WP4 - Invited SessionDate : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Kingfisher

Networked SystemsChairs : Ling SHI

Hong Kong University of Scienceand Technology

WP4.1 P0403 1620 - 1640Sensor Scheduling with LimitedCommunication Energy and BandwidthLing SHI, *Peng CHENG, *Jiming CHENCalifornia Institute of Technology, Pasadena*Zhejiang University

In this paper, we consider the problem of sensorscheduling with limited resources. Two sensors areused to measure the state of a discrete-time linearprocess. We assume that each sensor has amaximum duty cycle and at most one sensor cancommunicate with a remote estimator at each timestep due to the limited communication bandwidth.When a sensor is scheduled to send data, it sendsthe most recent D measurement data to theestimator. Upon receiving the measurement datafrom the sensors, the estimator computes the optimalestimate of the state of the process. We first presenta necessary condition for a sensor scheduling schemeto be optimal. Based on this necessary condition, weconstruct an optimal scheduling scheme thatminimizes the estimation error at the estimator andat the same time satisfies the energy andcommunication bandwidth constraints. We alsoprovide a sufficient condition on the minimum D suchthat an optimal scheduling scheme can beconstructed. Examples are provided throughout thepaper to demonstrate the results developed.

WP4.2 P0434 1640 - 1700Feedback Stabilization over StochasticMultiplicative Input Channels: Continuous-Time CaseNan XIAO, Lihua XIENanyang Technological University

The present work fits within the general study ofcommunication and control co-design, where themean square stabilization of continuous-timenetworked control systems over stochasticmultiplicative input channels is addressed. Motivatedby the limited communication capacity and networkresource allocation in multiple channelcommunication systems, we assume that the overallquality of service defined in this paper is fixed andcan be assigned among the input channels. We showthat there exists a minimal requirement on the overallquality of service for achieving the mean squarestabilization of the networked system. For the case ofstatic state feedback, a tight lower bound on theoverall quality of service for mean square stabilization

is derived in terms of the instability degree of theplant. In the case of output feedback, additionallimitations are induced by nonminimum phase zeros,where both stabilization over a single-input channeland stabilization of essentially triangular plants overmulti-input channels are studied. The application ofthe results to vehicle platooning is demonstrated.

WP4.3 P0483 1700 - 1720Graphic Interpretations of StructuralControllability for Switched Linear SystemsXiaomeng LIU, Hai LIN, Benmei CHENNational University of Singapore

This paper considers the controllability problem forswitched linear systems. In particular, the structuralcontrollability of switched linear systems isinvestigated. The structural controllability of switchedlinear systems is a generalization of the traditionalcontrollability concept for dynamical systems, andpurely based on the graphic topologies among stateand input nodes. First, two kinds of graphicrepresentations of switched linear systems areproposed. Second, several graphtheoreticcharacterizations of the structural controllability forswitched linear systems are presented based onthese two newly introduced graphs. Finally, the paperconcludes with several illustrative examples anddiscussions of the results and future work.

WP4.4 P0825 1720 - 1740Particle Filter Based Human Motion TrackingZhenning LI, Dana KULICUniversity of Waterloo

This paper proposes a particle filter based marker-less upper body motion capture system, capable ofrunning in realtime. This system is designed for ahumanoid robot application, and thus a monocularimage sequence is used as input. We first set up amodel of the human body, a sub-model whichincludes 11 Degrees of Freedom is used for the upperbody tracking. Considering the realtime processingrequirements, two time efficient cues areimplemented in the likelihood calculation, namely theedge cue and the distance cue. The system is testedusing a publicly available database, which consists ofboth the videos and the ground truth data, enablingquantitative error analysis. The system successfullytracks the human through arbitrary upper bodymotion at 20Hz.

WP4.5 P1069 1740 - 1800Convergence and Mean Square Stability ofOptimal Estimators for Systems withMeasurement Packet DroppingHuanshui ZHANG, Xinmin SONG, *Ling SHIShandong University*Hong Kong University of Science and Technology

This paper is concerned with estimation problemfor discrete-time systems with packet dropping. Anew optimal filter is derived by minimizing the mean

48

squared estimation error. An optimal smoother is alsoderived in a similar way. Both estimators aredesigned by solving one deterministic Riccatiequation. Both the convergence of the estimationerror covariance and mean square stability of theestimator are proved under standard assumption. Itis shown that the new estimator has smaller errorcovariance and has wider applications as comparedwith the MMSE estimator. One of the key techniquesadopted in this paper is the introduction of theinnovation sequence for the multiplicative noisesystems.

Session WP5Date : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Nightingale

Search, Rescue and Field RoboticsChairs : Jose M GIRON-SIERRA

Universidad Complutense de MadridVijay S RAJPUROHITGogte Institute of TechnologyBelgaum

WP5.1 P0265 1620 - 1640Autonomous Marine Craft Navigation: On theStudy of Radar Obstacle DetectionChiemela ONUNKA, *Glen BRIGHTUniversity of Kwazulu Natal*University of kwazulu Natal

Technological advancements over the years haveincreased the use of Radar technology in the field ofrobotics, especially in marine robotics to aid obstacledetection algorithms. Obstacle detection comprises ofan analytical process in which different algorithms areapplied to the field of study to determine the range ofobjects that are within the reach of a robot. Radarsignal analysis and target detection in conjunctionwith target tracking are attributes required forautonomous marine navigation. The paper presents amodel which converts optimal estimates of radarrange values for each range spectra into multipletargets down-range and also presents an approachfor power range spectra (range bins) prediction usingthe radar range equation with adequate informationof the signal-to-noise ratio (SNR) of the radar.Obstacle detection in the presence of noise raisescertain probabilities of false alarms. Targetcharacteristics are simulated and these arefluctuating targets and non fluctuating targets.Analytical models, simulations and techniques ofobstacle detection for autonomous marine craftnavigation using a continuous wave radar systemwere points of discussion in the paper.

WP5.2 P0423 1640 - 1700A Development Project of Autonomous MarineSurface Vehicles for Sea DeminingFernando J PEREDA, Hector GARCIA DE MARINA,Juan Francisco JIMENEZ, Jose M GIRON-SIERRAUniversidad Complutense de Madrid

a sea demining system using autonomous marinesurface vehicles (AMSV) is introduced. The researchinvolves the development of exemplars of thesevehicles, and the procedures for area scanning andcoverage. The demining is made by field influence,towing a submerged “fish”. This study is made bothwith simulations and with scale experiments

WP5.3 P0589 1700 - 1720Single-Human Multiple-Robot Systems for

49

Urban Search and Rescue - Justifications,Design and TestingChoon Yue WONG, Gerald SEET, Siang Kok SIM, WeeChing PANGNanyang Technological University

The evolution of interactions between humans androbots has reached a stage close to where multiplerobots can be robustly deployed. The domain ofUrban Search and Rescue stands to benefit from sucha capability. However, even autonomous robots mustbe supervised by humans. Yet, reasons such aseconomy, efficiency, added reliability as well asimproving overall group capability and addressingpayload constraints, make multiple robot deployment(MRD) an appealing possibility. Single-HumanMultiple Robot Systems facilitate MRD while ensuringthe supervision of robots deployed. However, thedesign of the single-human multiple-robot systemrequires consideration of key design attributes suchas collective size, composition and structure as wellas of the MRD issues of communication costs,computational costs, cognitive workload and situationawareness. The experiments to determine the effectsof the key design attributes on the issues of MRDhave been designed and are described in this paperas well.

WP5.4 P0602 1720 - 1740A Motion Controller for a Pan-tilt Camera on anAutonomous HelicopterHaibo DENG, Xiaoguang ZHAO, Zengguang HOUChinese Academy of Sciences

In this paper, a motion controller for a pan-tiltcamera mounted on an autonomous helicopter ispresented. The motion planner is designed accordingto the kinematics of the pan-tilt camera. However,the solution of the inverse kinematics of the pan-tiltcamera is not unique. To deal with this situation, adecision maker is designed. The decision makermakes use of a cost function to decide which solutionshould be chose. And the error signal is defined asthe difference between pan-tilt joint angles and thedesired joint angles. The dynamics of the errorsystem is derived, and proportional controllers aredesigned according to the error dynamics to controlthe pan and tilt joints respectively. This system isvalidated in a virtual reality environment, and thesimulation results show that this system can track thetarget successfully.

WP5.5 P0951 1740 - 1800The Use of Area Extended Particle SwarmOptimization (AEPSO) in Swarm RoboticsAdham ATYABI, David M W POWERSFlinders University

Swarm Robotics is the study of simple, un-intelligent robots teaming up together to addresscomplicated tasks using cooperation andknowledge/skills sharing factors. Particle SwarmOptimization (PSO) is an Evolutionary algorithminspired by animals’ social behaviors. PSO has beenused in various problems due to its fast convergence

capability. Area Extended PSO (AEPSO) is anenhanced version of PSO designed to addresscomplications in the Swarm Robotics field. Thesecomplications include dynamicity of the environment,degree of cooperation, time dependency of the tasks,and uncertain nature of the environment. This studyinvestigates advantages and shortcomings of theAEPSO method in the robotic domain.

50

Session WP6Date : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Pelican

Mobile Robotics IIChairs : Sai Hong TANG

University Putra Malaysia

WP6.1 P0300 1620 - 1640Towards A Fully-Autonomous Vision-basedVehicle Navigation System in OutdoorEnvironmentsPeyman MOGHADAM, Wijerupage Sardha WIJESOMA,Moratuwage M.D.P.Nanyang Technological University

Colour Stereo visions are the primary perceptionsystem of the most Unmanned Ground Vehicles(UGVs), which can provide not only 3D perception ofthe terrain but also its colour and texture information.The downside with present stereo vision technologiesand processing algorithms is that they are limited bythe cameras’ field of view and maximum range,which causes the vehicles to get caught in cul-de-sacs. The philosophy underlying the proposedframework in this paper is to use the near-field stereovision information associated with the terrainappearance to train a classifier to classify the far-fieldterrain well beyond the stereo range for eachincoming image. We propose an online, self-supervised learning method to learn far-field terraintraversability with the ability to adapt to unknownenvironments without using hand-labelled trainingdata. The method described in this paper enhancescurrent near-to-far learning techniques by automatingthe task of selecting which learning strategy to beused from among several strategies based on thenature of the incoming real-time input training data.Promising results obtained using real datasets fromthe DARPA-LAGR program is presented and theperformance is evaluated using hand-labelled groundtruth.

WP6.2 P1013 1640 - 1700Jumping Patterns Analysis for 1-DOF Two-legged RobotTaku SENOO, Yuichi TANNO, Masatoshi ISHIKAWAUniversity of Tokyo

In this paper, a jumping motion is considered withthe goal of achieving high-speed dynamic motion forlegged robots. We propose a two-step jumpingpattern which makes use of the characteristic of alightweight and fast actuator. In addition, jumpingstrategies based on the dynamics of the robots arepresented, and the effect of structural parameter onjumping height is analyzed. Experimental results areshown in which a 1-DOF two-legged robot jumps withthree patterns.

WP6.3 P0347 1700 - 1720Complete Coverage Algorithm based on LinkedSmooth Spiral Paths for Mobile RobotsTae-Kyeong LEE, Sang-Hoon BAEK, Se-Young OH,*Young-Ho CHOIPohang University of Science and Technology*Pohang Institute of Intelligent Robotics

This paper presents an on-line complete-coveragepath planning algorithm for mobile robots based onapproximate cellular decomposition, which abstractsthe target environment using grid. Most existing grid-based coverage algorithms have a common problemof constrained mobility which degrades the efficiencyof the coverage task by inducing zigzag like path. Inthis paper, we propose a new complete coveragepath generation algorithm, linked-smooth-spiral-path(LSSP), which removes the constraint on mobility byadopting a high-resolution grid-map representation ofthe environment and a cardinal-spline curve-model togenerate a smooth spiral coverage path. We define anew performance measure for coverage tasks, whichquantifies the smoothness of the coverage path.Simulation results demonstrate the improvedcoverage performance of the proposed algorithmcompared to other existing grid-based coveragealgorithms.

WP6.4 P0780 1720 - 1740Neuro-adaptive Virtual Leader BasedFormation Control of Multi-Unmanned GroundVehiclesZan YAO, Yongduan SONG, Wenchuan CAIBeijing Jiaotong University

The problem of formation control of multipleunmanned ground vehicles (UGVs) is studied in thispaper. The underlying problem is essentially a highdimensional multi-vehicle trajectory tracking controlproblem, its complexity grows significantly as thenumber of the involved vehicles increases. Toaddress the singularity inherent in regular formationprocess and the complexity involved in formation ofmultiple vehicles, this paper attempts an approachthat integrates robust adaptive neural network (NN)control with the virtual leader concept. Thefundamental idea behind this approach is to use thevirtual leader-follower format to convert the probleminto the one that only involves each individual vehicleto track the virtual leader. By doing so, the singularityassociated with the traditional formation control isalso avoided. As the system involves significantnonlinearities and uncertainties, robust adaptive NNcontrol algorithms are developed, which are shown tobe able to achieve high precision formation. Thefeasibility and effectiveness of the proposed methodare also verified by simulation studies.

WP6.5 P0272 1740 - 1800Survey Results on Status, Needs andPerspectives for using Mobile Service Robots inIndustrial ApplicationsAndreas KROLL, Samuel SOLDANUniversity of Kassel

51

Mobile robots for carrying out service tasks inindustrial plants and on infrastructure gainimportance e.g. due to aging assets anddemographics. In order to identify key areasrequiring research and development activities, aquestionnairebased online survey was conducted in2009. Addressed were participants working inindustry knowing best about needs, potentials,obstacles and pitfalls of applying mobile servicerobots in their companies/industrial sector. Thispaper presents the surveying methods and theobtained results which mainly cover chemical andautomotive industries due to their world-wideeconomic significance as well as their research andinnovation intensity.

Session WP7 - Invited SessionDate : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Oriole

Advances in Biometric Theory andApplications IIChairs : Massimo TISTARELLI

University of SassariWei Yun YAUInstitute for Infocomm Research

WP7.1 P0735 1620 - 1640Face Occlusion Detection by using B-splineActive Contour and Skin Color InformationGahyun KIM, Jae Kyu SUHR, Ho Gi JUNG, Jaihie KIMYonsei University

This paper proposes a face occlusion verificationmethod for an automated teller machine (ATM)application. The proposed method mainly consists ofthree steps. Firstly, a head and shoulder shape isdetected by applying B-spline active contour tomotion edges. This motion edge is generated by akurtosis-based frame selection and distancetransformationbased motion edge detection.Secondly, a face area is estimated by fitting an ellipseto the detected head and shoulder shape. Finally,occlusion of the face area is determined bymeasuring skin color area ratio (SCAR) of whole facearea and facial component areas. Experimentalresults show that the proposed head and shoulderdetection method has 94.8% detection rate eventhough there are various types of severe occlusionsin faces, and the proposed occlusion verifier has86.7% verification rate.

WP7.2 P1121 1640 - 1700Eye Feature Extraction using K-meansClustering for Low Illumination and Iris ColorVarietyVan Huan NGUYEN, Thi Hai Binh NGUYEN, Hakil KIMInha University

This paper presents an approach for locating eyefeatures in color images based on the unsupervisedK-means clustering. Given the assumption that theinput is an eye window containing a single eye, theproposed method detects the iris by unsupervised K-means clustering on the feature spaces ofcompensated red and green color channels. The iriscircle is then refined using the gradient informationand circular Hough transform. For the scleradetection, the r-g and r-b are utilized as they showthe discriminant feature of sclera regardless of lightcondition and iris color. The sclera is then extendedto fit the eyelids by a region growing scheme.Experiments on a collection of eye images extractedfrom FERET facial database and our self-collectedimages show a promising performance toward thelow illumination and iris color variety.

52

WP7.3 P1147 1700 - 1720A Comprehensive Sclera Image QualityMeasureZhi ZHOU, Yingzi, Eliza DU, N. Luke THOMASIndiana University-Purdue University Indianapolis

Sclera recognition can be used for humanidentification. However, if the sclera images cannotbe properly segmented by the system or the imagesof sclera patterns are defocused and/or saturated, itcan significantly affect the accuracy of sclerarecognition. In this paper, we propose acomprehensive sclera image quality measure whichcan quickly detect if the image has a valid eye ,assess the image quality, evaluate the segmentationaccuracy, and measure if the image has sufficientfeature information for recognition. In addition, itused Dempster Shafer Theory to fuse the qualityscore, segmentation score, and feature scoretogether to generate the overall combination score. Itis empirically verified using the UBIRIS database thatthe proposed quality measure is highly correlatedwith the performance of sclera recognition.

WP7.4 P0991 1720 - 1740Effects of Facial Alignment for Age EstimationHee Lin WANG, Jian-Gang WANG, Wei Yun YAU,*Xing Lun CHUA, *Yap-Peng TANInstitute for Infocomm Research*Nanyang Technological University

Age estimation is an important enabling capabilityfor the near future, especially in applications relatedto Human Computer Interaction. Perhaps due totechnical difficulties, age estimation has only recentlybegun to receive more attention. One of the moreimportant pre-processing steps before age estimationis facial alignment, which spatially transforms a faceimage to align certain facial features, in order tomaximize classification accuracy. However thecapability for facial alignment in automated ageestimation literature is commonly assumed, and theeffects of facial alignment is not addressed, eventhough it is an important issue especially for livedeployment. In this paper, we present, to ourknowledge, the results of the first systematicinvestigation on the effects of facial alignment on ageestimation accuracy, and conclude with somedirections on further topics of investigation in thisarea.

WP7.5 P1148 1740 - 1800Speaker Gender Recognition using Score LevelFusion by AdaBoostMasatsugu ICHINO, Naohisa KOMATSU, *Jian-GangWANG, *Wei Yun YAUWaseda University*Institute for Infocomm Research

We propose speaker gender recognition achievedby using score level fusion by AdaBoost. Softbiometrics has been focused on because recognitionby fusing biometric systems and soft biometric traits

may improve the accuracy of recognition anddecrease the time for this. Gender recognition isimportant for speaker recognition and can provideimportant information to speaker recognitionsystems. Mel-frequency cepstral coefficient (MFCC)and pitch contain gender information. MFCCs andpitch are often used for gender recognition.Consequently, identification accuracy may beimproved by using both MFCC and pitch. We focusedon the score level fusion to accomplish speakergender recognition. We propose speaker genderrecognition based on the score level fusion usingAdaBoost because it can control the recognitionaccuracy and recognition time. We experimentallydemonstrate the proposed method's effectivenessthrough simulation results and show that it achievesgreater accuracy than that obtained by using singleinformation from voice.

53

Session WP8 - Invited SessionDate : Wednesday, 08 December 2010Time : 1620 - 1800Venue : Bluebird

Advanced Concept for Emerging ComputerVisionChairs : Siu Yeung CHO

Nanyang Technological University

WP8.1 P1110 1620 - 1640Two-tier Emergent Self-Organizing (TtEsom)Approach of Understanding EmotionsYok Yen NGUWI, *Teik Toe TEOHJames Cook University Australia, Singapore campus*Nanyang Technological University

This paper extends the previous work on emotionmapping [1] that attempts to emulate human brainreference model. Most emotion recognition systemanalyzes facial expression through supervisedlearning whereas this work adopts unsupervisedlearning. The system first locates the human face inan image, and then identifies the localized faceemotion. The proposed method uses featuresobtained using Gabor wavelets, undergoes featuresselection through the use of a derivation of SupportVector Machine. This work adopted a connectionistmodel, called Two-tier Emergent Self-Organizing Map(TtEsom) to analyse the emotion. The result showsimprovement over the previous work and comparableresult with supervised learning approach.

WP8.2 P1106 1640 - 1700A Segmentation based Approach for ShapeRecovery from Multi-color ImagesM. K. M. RAHMAN, *Tommy W. S. CHOW, **SiuYeung CHOUnited International University*City University of Hong Kong**Nanyang Technological University

Conventional shape from shading (SFS) algorithmsare unable to deal with multi-color image satisfactory.This is because the assumption of constant surfacealbedo in the algorithms is not applicable to multi-color images. This paper proposes a new SFSapproach for multi-color images through asegmentation-based shading recovery technique.With this technique a gray image is firstly extractedfrom the multi-color image containing better shadinginformation compared with other color-to-grayconversion methods. The shading is recovered in thegray image as if the objects were made of singlecolor. Shape of the multi-color object can then berecovered by classical gray-scaled SFS methods.Experimental results with synthetic and real multi-color images are presented. The obtained resultscorroborate that the proposed scheme is able todeliver better performance compared with other colorSFS methods.

WP8.3 P1108 1700 - 1720A Color-based Approach for DisparityRefinementRaj Kumar GUPTA, Siu Yeung CHONanyang Technological University

We present a new disparity refinement algorithmthat utilize color information of the reference imageto generate sharp disparity maps. While existingmethods use iterative approaches or require multipledisparity maps, the proposed algorithm uses a singlepass approach to reduce errors at depthdiscontinuities. The experimental results areevaluated on the Middlebury benchmark dataset;show the effectiveness of the proposed algorithm.

WP8.4 P1111 1720 - 1740Emotion Indexing using Hidden Markov ExpertRule Model (HMER) for Autism ChildrenTeik Toe TEOH, *Yok Yen NGUWINanyang Technological University*James Cook University Australia, Singapore campus

Hidden Markov Models (HHMs) have been appliedsuccessfully in the field of applied sciences andengineering [1]. The potential applications inmanufacturing industries have not yet been fullyexploited. In this paper, we propose a Hidden MarkovExpert Rule Model (HMER) to index emotion as partof the assistive technology (AT). We propose to index4 emotions: neutral, happy, sad and surprise.Numerical examples are given to illustrate theeffectiveness of the proposed models. HMER is a partof AT that can be used to increase, maintain, orimprove functional capabilities of individuals withdifficulties in recognizing emotions. It promotesgreater independence for this group of people byenabling them to perform task that they wereformerly unable to accomplish. Children with autismspectrum disordered (ASD) have difficulty recognizingemotions in themselves and others. This workpresents a fast cognitive assistive Hidden Markov-based emotional indexer which can help children withASD to read and respond to the facial expressions ofpeople they interacting with. The result of emotionindexer is very encouraging; it achieves accuracy ofabout 70% and the respond time is around 2 framesper seconds

WP8.5 P1114 1740 - 1800Fast Polar Harmonic TransformsZhuo YANG, Sei-ichiro KAMATAWaseda University

Polar Harmonic Transform (PHT) is termed torepresent a set of transforms those kernels are basicwaves and harmonic in nature. PHTs consist of PolarComplex Exponential Transform (PCET), Polar CosineTransform (PCT) and Polar Sine Transform (PST).They are proposed to represent invariant imagepatterns for two dimensional image retrieval andpattern recognition tasks. They are demonstrated toshow superiorities comparing with other methods ondescribing rotation invariant patterns for images.

54

Kernel computation of PHTs is also simple and has nonumerical stability issue. However in order toincrease the computation speed, fast computationmethod is needed especially for real worldapplications like limited computing environments,large image databases and realtime systems. Thispaper presents Fast Polar Harmonic Transforms(FPHTs) including Fast Polar Complex ExponentialTransform (FPCET), Fast Polar Cosine Transform(FPCT) and Fast Polar Sine Transform (FPST) that arededuced based on mathematical properties oftrigonometric functions. The proposed FPHTs areaveragely over 6~8 times faster than PHTs thatsignificantly boost computation process. Theexperimental results on both synthetic and real dataare given to illustrate the effectiveness of theproposed fast transforms.

Session WIDate : Wednesday, 08 December 2010Time : 1015 - 1800Venue : Foyer

Interactive Session I

WI.1 P0309Development and Application of AutomationControl System to Plate Production LineZhi-jie JIAO, Chun-yu HE, Jun WANG, Zhong ZHAONortheastern University

The automation control system to plate productionline is developed independently. The whole system isdivided into several levels and control functions areput into effect by the basic automation system,process automation system and human-machine-interface (HMI) system. Plate production thicknessaccuracy depends on the Automatic Gauge Control(AGC) function in the basic automation system for theplate mill area. With date communication module,process tracking module, data management module,and model calculation module, the core processmodel setup function is implemented by processautomation system. Process monitoring and operationare implemented by HMI system. This automationcontrol system proposed here has been appliedsuccessfully to several plate production lines in Chinawith a good effect.

WI.2 P0340Passive Tracking using TDOA for Super-low-altitude TargetPanlong WU, Yuming BO, Jianshou KONG, Xingxiu LINanjing University of Science and Technology

A novel near-space floating platform aided passivelocation method for super-low-altitude target usingtime difference of arrival(TDOA) is proposed. Thenew method uses a near-space floating platform asthe baseline in the height to form a redundant1ocation system consisting of several subsystems.Each subsystem can obtain a set of location resultswith ambiguity. After eliminating the ambiguity by thenearest matching, the new method can improve thelocation accuracy in the whole observed airspace bysimplified weighted least square(SWLS) fusion. Abetter tracking performance can be achieved bytreating the SWLS fusion location result as pseudo-linear measurement, and keeping the trackingthrough Kalman filter. Simulation results show thatthe proposed method comes to stabilization within40s and has higher tracking precision than the SWLSmethod.

WI.3 P0414Improvements of Flashing Mode byProgrammed Control for Multiple Colors in LEDOptical Stimulus SystemKen ISHII, Go TAKAYAMA, *Hiroshi INADA,

55

**Tomohiro MIKI, ***Makoto NARUMI, ***YasunoriSAKURAIFisheries Research Agency*Tokyo University of Marine Science and Technology**Towa Denki Seisakusho Co., Ltd.***Hokkaido University

Behavior observation examination of Japaneseflying squid against optical stimulus is scheduled. ALED optical resource system has been designed as anoptical stimulus for this examination. Two kinds of aLED driver and panel system were developed. Oneconsists of constant current power supply units. Theother is a LED driver based on Pulse WidthModulation (PWM). About the former, a control andmonitor program via a personal computer (PC) wasdeveloped. In this paper, the former was described indetail. By control of two parameters of preset currentand voltage to a constant current power supply unit,a waveform of LED current was improved. Leading-up time is shorter than nominal value. Overshoot wasimproved in comparison with a MOSFET switchcontrol. Flashing worked in LED current higher thanrated current value.

WI.4 P0671Artificial Neural Network Based E-Nose andtheir Analytical Applications in Various FieldMaria JAMAL, *M R KHAN, *Saiyad A IMAM, **ArifJAMALGGS Indraprastha University*Jamia Nagar,Okhla, New Delhi**CPWD

A brief and historical overview of research anddevelopment in the field of artificial neural networkbased electronic nose system is presented.Electronic-nose devices have received considerableattention in the field of sensor technology during thepast twenty years, largely due to the discovery ofnumerous applications derived from research indiverse fields of applied sciences. Electronic/artificialnoses are being developed as systems for theautomated detection and classification of odors,vapors, and gases. An electronic nose is generallycomposed of a chemical sensing system (e.g., sensorarray or spectrometer) and a pattern recognitionsystem (e.g., artificial neural network). We aredeveloping electronic noses for the automatedidentification of volatile chemicals for environmental,medical applications, commercial industries, includingthe agricultural, biomedical, cosmetics, food,manufacturing, military, pharmaceutical, regulatoryapplications and various scientific research fields. Thispaper is a review of the major electronic nosetechnologies, developed since this specialized fieldwas born and became prominent in the mid 1980s,and a summarization of some of the more importantand useful applications that have been of greatestbenefit to man.

WI.5 P0767Dynamic Fuzzy Neural Network BasedPredictive Control for Alternating Current

Excitation GeneratorsZhifei ZHANG, *Xuan WANGFoshan University of Science and Technology*Guangdong University of Technology

Alternating current excitation generators (ACEG)can adjust the active power and inactive powerflexibly and improve the stability of power system.The key to enhance the power system’s stability is tochoose appropriate ACEG’s excitation control method.Con-ventional excitation controllers are unable toperform optimally over the full range of operationconditions and disturbances, due to the highlycomplex, non-linear nature of power systems. In thispaper, dynamic fuzzy neural network based predictivecontrol is proposed to cope with the problem. Fuzzyneural network is employed to predict power angleand stator voltage of ACEG excitation control system,in order to achieve good dynamics of fuzzy neuralnetwork, genetic algorithm is introduced to optimizenetwork parameters. Based on the model output,branch-and-bound optimization method is adopted,which generates proper value of excitation controlvariable of ACEG. Fuzzy neural network based modelpredictive algorithm is used in internal model controlscheme to compensate for process disturbances,measurement noise and modeling errors. Simulationtest under large disturbance at various operatingpoints is made. The results show the controller iseffectiveand feasible.

WI.6 P0982Equivalence and Stability of Two-layer CellularNeural Network Solving Saint Venant 1DEquationVu Duc THAI, *Pham Thuong CATThai Nguyen University*Vietnam Academy of Science and Technology

Cellular Neural Network (CNN) has been used forsolving Partial Differential Equations (PDE). However,the equivalence and stability of system should beconsidered carefully in a particular problem. In thispaper, we introduce the model CNN for solving set oftwo PDEs describing water flow channels (called SaintVenant equation). We analyze the approximation andtopological equivalence issues between CellularPartial Difference Differential Equation (CPDDE) andits original PDEs. The stability of CNN system is alsoproved from discovering the equilibrium of the stateand output of each cell. The paper has 4 parts. Afterintroduction, part 2 gives a two-layered CNN 1Dmodel for solving PDE Saint Venant equation. In thepart 3 the equivalence and stability of the CNN modelare proved, then simulation using FPGA. Theconclusions are given in the last part

WI.7 P0361Interval Type-2 Non-Singleton Type-2 Takagi-Sugeno-Kang Fuzzy Logic Systems using theHybrid Learning Mechanism Recursive-Least-Square and Back-Propagation MethodsGerardo M. MENDEZ, Maria de los AngelesHERNANDEZ

56

Instituto Tecnologico de Nuevo LeonThis article presents a novel learning methodology

based on the hybrid mechanism for training aninterval type-2 non-singleton type-2 Takagi-Sugeno-Kang fuzzy logic systems (FLS). Using input-outputdata pairs during the forward pass of the training andprediction processes, the interval type-2 nonsingletontype-2 TSK FLS the consequent parameters weretuned by using the recursive least squares (RLS)method. In the backward pass, the antecedentparameters were tuned by using the back-propagation (BP) method. As reported in theliterature, the performance indexes of these hybridmodels have proved to be better than the individualtraining mechanism when used alone. The proposedhybrid methodology was tested thru the modelingand prediction of the steel strip temperature at thedescaler box entry as rolled in an industrial hot stripmill. Results show that the proposed methodcompensates better for uncertain measurements thanprevious type-2 Takagi-Sugeno-Kang using non-hybrid or only back propagation learningmechanisms.

WI.8 P1053A Research on Parabolic Trough Solar CollectorSystem Control based on Hedge AlgebraNguyen Huu CONG, Ngo Kien TRUNG, Nguyen TienDUY, Le Thi Thu HAThai Nguyen University of Technology

This paper presents a new method in controlling aparabolic trough solar collector system to improve theefficiency of the solar-to-thermal energy. It isdesigning an intelligent pre-processor using HedgeAlgebra algorithm to calculate the setpoint for thecontrol loop, in which besides the information aboutthe trajectory calculated by a software, wind speedand fluid temperature of the collector are included.Moreover, this paper introduces a new simple flexiblecalculation tool which enables to calculate with ahigher accuracy.

WI.9 P0185A Study on Tracking Position Control of anPneumatic System by Backstepping DesignChia-Hua LU, Yean-Ren HWANGNational Central University

The pneumatic actuator is widely used in industrialapplications due to its reliability, low cost and nooverheat problem in an industrial servo system.Recently, the development of control technology isimproving and the requirement for control precisiongets higher than before. In order to accomplishaccurately control performance, nonlinear controltechniques are implemented on control system. Thispaper presents a new form of backstepping slidingmode controller for X-Y table motion system.Experimental results are presented to show that theproposed controller can accomplish accurate trackingcircle trajectory performance

WI.10 P0558Reactive Power-Voltage Integrated ControlMethod based on MCRJin TAO, Jun WANG, Li WAN, *Dan CHEN,*Xiaosheng HUANG, **Wenjun ZENGElectrical Power Test and Research Institute of Hubei*Nanning Power Supply Company**Wuhan University

In power system, the voltage pulsation anddeficiency or surplus of reactive power will have agreat effect on the quality of electric power supply.Through the appropriate and effective way ofcompensation, we can keep the balance of reactivepower in the maximum degree by the maintenance ofacceptable voltage. That’s to say, quality of powersupply can be improved greatly by the syntheticcompensation of reactive power and voltage. Thispaper firstly describes the principles of syntheticcompensation of reactive power and voltage, thenintroduces the analyzes the principle of magneticallycontrolled reactor-MCR in detail and derives thecharacteristics, lastly designs a controlling strategy ofreactive power and voltage based on MCR, so as toachieve the goal of dynamic reactive powercompensation.

WI.11 P0753A Generic Procedure for Troubleshooting ofPLC Based Control SystemsArshad MALIK, *Muhammad CHOUDHRY, *TahirMEHMOOD, *Aamir HANIFGhazi Brotha Power House Attock*University of Engineering and Technology, Taxila

This paper describes the problems generally facedwith PLC based control systems and a generalprocedure to solve them. Flowcharts fortroubleshooting of power supply faults, internal andexternal problems of PLC has been included in thepaper. The approach mentioned for troubleshootingis general and can be applied on any type of PLCbased control system. A case study is included tovalidate the developed general procedure for thetrouble shooting of PLC based controlled system.

WI.12 P0835Impulsive Disturbance Rejection in Hard DiskDrivesTingting GAO, *Chunling DU, Lihua XIE, Wen-JianCAINanyang Technological University*Data Storage Institute

This paper proposes filtering methods to cancelthe impulsive disturbance contained in position errorsignal (PES) in hard disk drives (HDDs). Theimpulsive disturbances may be observed as a fewsingle sudden changes or some consecutive changesin PES. The filtering includes impulsive disturbanceidentification and estimation. A recursive method isused to determine the dynamic boundaries foridentification. Two methods are proposed forestimation: a linear interpolation and an adaptive

57

least mean square (LMS) algorithm. The former isused to estimate the normal PES directly and thelater is used to estimate the impulsive disturbance forcancellation. It turns out that these methods are ableto effectively cancel the impulsive disturbance and donot affect the servo performance.

WI.13 P0874Optimal Control for EHA-VPVM System basedon Feedback Linearization TheoryYe ZHANG, Yongling FU, Weixing ZHOUBeihang University

In this paper, an optimal control method based onfeedback linearization is proposed for EHA-VPVM(electro-hydrostatic actuator with variable pumpdisplacement and variable motor speed) system,which is a dual-input single-output (DISO) nonlinearsystem with multiplicative nonlinear property. In thiswork, a real EHA-VPVM system is modeled at first.Then the nonlinear model is linearized throughfeedback linearization and a control method isdesigned by using optimal control theory based onthe linearized model. Simulation results show that ourcontrol method is effective for EHA-VPVM positioncontrol system, and can suppress perturbation ofexternal load force to some degree. In addition, bysimply adjusting the weighting coefficients of thedisplacement and the rotational speed of pump, thismethod can meet the different requirements ondynamic response and energy loss of system.

WI.14 P0884Subspace Identification and Model PredictiveControl for BuildingsJiri CIGLER, Samuel PRIVARACzech Technical University in Prague

Model predictive controller presented in this articlemakes use of both weather forecast and thermalmodel of a building to control inside temperature.This, by sharp contrast to conventional controlstrategies such as weather-compensated heatingcontrol (heating-curve controller), enables utilizationof thermal capacity of the building, thus minimizationof energy consumption. The inside temperature canbe maintained at desired levels independent of theoutside weather conditions using modifiedformulation of predictive controller. Nevertheless,proper identification of the building model is crucial.The models of multiple-input multiple-output systemscan be identified using subspace methods. Thecontroller was tested on (and applied to) the realbuilding and results were compared with a presentheating control.

WI.15 P0960Modeling Method based on Signal Flow GraphCorrelation Matrix and its ApplicationBingxin HAN, Lixian LIU, Zhanfeng GAO, Liqiang DUShijiazhuang Tiedao University

This paper presents a new modeling method

based on signal flow graph correlation matrix. Firstly,according to the signal flow graph, a signal tree canbe selected. Based on the signal tree, a signal branchand the node can be added, finding independent loop- loop incidence matrix. The correlation matrix candetermine the relevance of the loop, and help seekits loop gain. Moreover, mathematical model can befigured out by using Mason formula. Theoreticalanalysis and experimental results show that when thenodes increase, this method is simpler and themeaning of the formula is clearer. In comparison toexisting methods, our method has obviousadvantages in calculation and expression. Ourmethod makes the model of the system moreaccurate, and the modeling process becomes fasterand more convenient.

WI.16 P1080Laboratory Demonstration for Model PredictiveMultivariable Control with a Coupled DriveSystemSteven Weidong SU, Hung T NGUYEN, Quang P HAUniversity of Technology, Sydney

Teaching multivariable control usually involves acertain level of mathematical sophistication andhence requires some labaratorial exemplification ofthe material given in formal lectures. This paperreports on a hands-on approach to multivariablecontrol education via the implementation of a modelpredictive controller on a two-input, two outputcoupled drive apparatus. This scaled-down systemrepresents many industrial processes while providesan excellent set-up for demonstrating the cross-coupled effects in multi-input multi-output systems.Here, a model predictive controller (MPC) isdeveloped and implemented on the basis of aconstrained optimization problem to show controlperformance via the belt tension and velocityoutputs, demonstrate the decoupling capability, andalso illustrate such issues as control input saturation,the selection of operating point, reference inputs, andsystem robustness to external disturbance andvarying parameters. The implementation is based onLabview and MATLAB Model Predictive ControlToolbox.

WI.17 P0696Stability Analysis of Practical AnisotropicSwarmsLiangshun WANG, Huajing FANGHuazhong University of Science and Technology

This paper considers finite sense ability, finitemotion ability and anisotropy into swarm system. Wefirst construct an anisotropic model with a class ofattraction and repulsions in terms of finite senseability. Then we analyze their stability. It is shownthat the individuals of the swarm model aggregateand eventually enter into a bounded region aroundthe swarm center. This study is more applicable topractical engineering.

58

WI.18 P0793Instantaneous Availability Model with 2-DDiscrete Characteristics of One-unit RepairableSystemsYi YANG, *Yongli YU, Lichao WANG, Rui KANGBeihang University*Ordnance Engineering College

The discrete-time one-unit repairable system wasstudied, whose lifetime, repair delay time, correctivemaintenance time and preventive maintenance timeare all assumed to be random variables with generaldiscrete distributions. We investigated therelationship of system states including operationalstate, waiting state for repair, state for preventivemaintenance, and state for corrective maintenance.Then the state transition model was built.Furthermore, we established the instantaneousavailability model which has the 2-D discretecharacteristics, and on the basis, we proposed theoptimal maintenance interval model. Finally,numerical examples were given to illustrate theproposed models.

WI.19 P0891Design of Signal Control Structures usingFormal Methods for Railway InterlockingSystemsOytun ERIS, Ilhan MUTLUIstanbul Technical Universty

Today the relay based train interlocking systemsare changing into programmable softwareinterlocking systems. Looking at the hardware aspect,safety certified components can be used in order tosatisfy the safety requirements of train interlockingsystems. But to satisfy the software requirements,the design and the programming of the interlockingsystem must be made by formal methods. Preparingfunction blocks for each kind of the field equipmentscan be a method in order to realize a railwayinterlocking system. The purpose of this paper is todevelop a PLC program for the signal function blocksby using formal methods and make a comparisonbetween them.

WI.20 P0621Haptic Microrobotic Intracellular InjectionAssistance using Virtual FixturesAli GHANBARI, XiaoQi CHEN, Wenhui WANG, *BenHORAN, *Hamid ABDI, *Saeid NAHAVANDIUniversity of Canterbury*Deakin University

In manual cell injection the operator reliescompletely on visual information for task feedbackand is subject to extended training times as well aspoor success rates and repeatability. From thisperspective, enhancing human-in-theloop intracellularinjection through haptic interaction offers significantbenefits. This paper outlines two haptic virtualfixtures aiming to assist the human operator whileperforming cell injection. The first haptic virtualfixture is a parabolic force field designed to assist the

operator in guiding the micropipette’s tip to a desiredpenetration point on the cell’s surface. The second isa planar virtual fixture which attempts to assist theoperator from moving the micropipette’s tip beyondthe deposition target location inside the cell.Preliminary results demonstrate the operation of thehaptically assisted microrobotic cell injection system.

WI.21 P0526Non-cooperative Target Assignment usingRegret MatchingShemin KALAM, Mahbub GANIKing's College London

In this paper, we have adopted a game theoreticapproach to consider optimal assignment of targetsto vehicles. We have taken optimality to mean thatthe cost of vehicle-target assignment is minimizedwhere the cost is the sum of the Euclidean distancesbetween a vehicle and its assigned target. Thisoccurs when the set of targets is assigned to theirnearest vehicles. In other words, each vehicle shouldassign themselves to a target located within theVoronoi region associated with the vehicle. Standardschemes require vehicle-vehicle communication atleast to rank the targets according to the nearestneighbour rule. In our proposed model, there is noneed for communication among the vehicles and onlylimited communication between the targets and thevehicles is required. Specifically, we have introducedan appropriate utility function which depends on thedistance between the vehicles and targets and thenumber of vehicles engaging a particular target. Thevehicles negotiate their choice of targets via regretmatching. We present simulations which demonstratethat vehicles select targets that fall within theirVoronoi region. For vehicles that do not have anytargets within their Voronoi region, they select theirnearest unassigned target. We also present analysisof how the designed utility function causesconvergence of the vehicle-target assignment usingregret matching.

WI.22 P05673D Hydrodynamic Analysis of BiomimeticRobot FishZhenying GUAN, Weimin GAO, Nong GU, SaeidNAHAVANDIDeakin University

This paper presents a three-dimensional (3D)computational fluid dynamic simulation of abiomimetic robot fish. Fluent and user-definedfunction (UDF) is used to define the movement of therobot fish and the Dynamic Mesh is used to mimic thefish swimming in water. Hydrodynamic analysis isdone in this paper too. The aim of this study is to getcomparative data about hydrodynamic properties ofthose guidelines to improve the design, remotecontrol and flexibility of the underwater robot fish.

WI.23 P05776 Degrees of Freedom (DOF) Maritime Robotic

59

Simulation FrameworkKuan Meng TAN, *Amir ANVAR, Tien-Fu LUThe University of Adelaide*Defence Science and Technology Organisation

Testing and gauging Autonomous UnderwaterVehicle (AUV) missions through observations andexperimental trials prove to be time consuming andcostly. A system capable of verifying the AUV controlmodels prior to practical test is needed. One of thekey aspects in the presented framework is AUVmodeling which mainly covers the area of AUVdynamics. Based on this designed simulationframework, custom scenarios provided by the usercan be modeled and its corresponding 6DOFdynamics can be observed. The simulation frameworkis built based on a developed AUV test-bed whichwas jointly upgraded by DSTO and the University ofAdelaide.

WI.24 P0896A Consistent Kinematic Modeling Method forMobile ManipulatorsJoono CHEONGKorea University

In this paper, a consistent kinematic modelingmethod for mobile manipulators is proposed. Wetransform the mobile manipulators into conventionalmanipulators via appropriate change of kinematicstructure. Consequently, analyses of kinematics formobile manipulators can be conveniently carried outwith the transformed manipulators. Cases ofdifferential-drive and car-like types of mobileplatforms are formulated, but theoretically not limitedto these.

WI.25 P0532Dynamic Analysis of a Flexure-basedMechanism for Precision Machining OperationYanling TIAN, Dawei ZHANG, *Bijan SHIRINZADEHTianjin University*Monash University, Clayton

This paper presents the dynamic modelling andperformance evaluation methodologies of a flexure-base mechanism for ultra-precision grindingoperations. The mechanical design of the mechanismis briefly described. A piezoelectric actuator is used todrive the moving platform. A flexure-based structureis utilized to guide the moving platform and toprovide preload for the piezoelectric actuator. Bysimplifying the Hertzian contact as a linear spring anddamping component, a bilinear dynamic model isdeveloped to investigate the dynamic characteristicsof the flexure-based mechanism. Based on theestablished model, the separation phenomenon ofthe moving platform from the piezoelectric actuator isanalyzed. The influence of the control voltage on themaximum overshoot is also investigated. The slopeand cycloidal command signals are used to reduceand/or avoid the overshoot of such flexure-basedmechanism under step command signal actuationcondition. The effects of the rising time of thecommand signals on the maximum overshoot and the

settling time are studied.

WI.26 P0748Parallel Implementation of 3D ObjectReconstruction in a Robotic NavigationalEnvironmentVijay S RAJPUROHIT, *Manohara PAI M MGogte Institute of Technology Belgaum*Manipal Institute of Technology, Manipal

Stereo vision systems in a Robotic navigationenvironment determine the depth in the form of adepth map image from two or more images whichare taken at the same time, but from slightly differentviewpoints. Ground suppression from depth map isessential for object reconstruction in 3D environmentas the ground surface is considered as travelingmedium than as an object. Ground separated depthmap needs to be segmented to identify the objects ofinterest. Objects identified below the threshold sizeneed to be removed from the scene. In this paper wepropose a Fuzzy-based parallel object clusteringalgorithm to reconstruct the objects of interest in 3Denvironment from the ground suppressed depth map.The depth map is divided into Fuzzy layers in X-Zplane and these Fuzzy layers are processed in parallelto reconstruct the objects of interest in 3Denvironment using region growing method on acluster computing setup. Experimental results showthat Fuzzy logic effectively handles the uncertainty inobject reconstruction process. The parallelimplementation of the algorithm drastically reducesthe time required for object reconstruction process incongested navigational environments. Index Terms—Stereo vision, Depth map, Parallel Depth mapGeneration, Cluster computing setup Scenesegmentation, Region growing algorithm, Fuzzynavigational environment.

WI.27 P1073Optimal Distribution of Propulsion for anAmphibious Robot based on Wheel-Propeller-Leg Mixed ThrustersYuangui TANG, Chongjie LIU, Aiqun ZHANG,Jiancheng YUChinese Academy of Sciences

A wheel-propeller-leg integrated amphibious robotwith multi-modes motion is presented as a new typeof robot, which can both crawl on land and swimunderwater in certain depth. The performance ofswimming with the mixed thrusters composed ofwheel-propellers and legs shows its propulsivecapability when the robot swims in water. Based onanalysis of propulsive principle of the wheel-propellerand the leg, hydrodynamic characters of the mixedthrusters are calculated, and the approximate modelsfor propulsion of the wheelpropeller and leg areobtained according to computational calculation andsimulation with commercial software. Considering thefeature of energy-taking by itself, the optimaldistribution of propulsion based on the wheel-propeller-leg mixed thrusters with the maximalnavigational distance is studied using multi-objective

60

optimization theory. As a result, the optimized drivingparameters for the mixed thrusters are achieved tobring the maximal propulsion while consuming theminimal energy. Experiments carried out in the poolprove the validity and rationality of the optimizationresults.

WI.28 P0692Random Force based Algorithm for LocalMinima Escape of Potential Field MethodJinseok LEE, Yunyoung NAM, Sangjin HONGStony Brook University

We address a new inherent limitation of potentialfield methods, which is symmetrically aligned robot-obstacle-goal (SAROG). The SAROG involves onecritical risk of local minima trap. For dealing with theproblem, we investigate the way how the localminima trap is recognized, and present our randomforce algorithm. The force algorithm has twocategories of random unit total force (RUTF) andrandom unit total force with repulsion removal(RUTF-RR) which are selected based on theconditions of a robot, an obstacle and a goal.

WI.29 P0784Development of Human-friendly Robot Armwith Adjustable Joint ComplianceShinya KAJIKAWA, Masashi NASUNO, KazutoHAYASAKATohoku Gakuin University

Safety is primary considered in designing a robotthat operates in human environment.This paperpresents a human-friendly robot arm designed so asto realize a physical interaction with human safely.This arm has a 3-D.O.F (shoulder joint : 2-D.O.F,elbow joint : 1-D.O.F). To guarantee the safety ofrobot motion, elbow joint has a multi-directionalpassive compliance.In this joint, two kinds of aircushions (Cushion A and Cushion B) are placedbetween a motor-driven disk and an output link. Therotational motion of the motor is transmitted to theoutput link with the friction force between the surfaceof Cushion A and the motor-driven disk. In addition,Cushion A plays following two important roles: (1)Absorption of external forces by its elasticdeformation and slippage on the motor-driven diskand (2) Sensing the directions and amplitude ofexternal forces with the pressure change in it. On theother hand, Cushion B controls the degree ofpressing Cushion A against the motor-driven disk byits expansion. The increment of pressing force ofCushion A results in the increment of joint stiffness,because the elastic deformation and slippage ofCushion A are prevented. This paper describes thestructure of the robot arm and the detail mechanismand performance of the proposed joint module.

WI.30 P1157Analysis of Real-time Velocity Compensationfor Outdoor Optical Mouse Sensor OdometryRobert ROSS, John DEVLIN

La Trobe UniversityThis paper investigates the linearity of optical flow

odometry with respect to velocity when usingrefocused optical mouse sensors for outdoor roboticnavigation. Optical mouse sensors are small,inexpensive and contactless devices which include alow resolution CMOS camera, DSP hardware andoptical flow firmware to provide optical displacementmeasurements in two dimensions. We performexperiments using different velocities, samplingperiods and with decoupled acceleration to develop alinear, real-time compensation algorithm providingvelocity corrected displacement measurements.

WI.31 P0394IVFH* : Real-time Dynamic Obstacle Avoidancefor Mobile RobotsJie DONG, Xueming MA, Kaixiang PENGUniversity of Science and Technology Beijing

VFH* could not make optimal choice of direction inthe narrow stability region. An improved methodcalled IVFH* enlarges certain value (CV), raise moregrids, add to relative threshold and introduce theconcept of reactive deformation links to improve theoriginal algorithm. IVFH* use Newtonian Physics andHooker’s Law to update the position of the nodes anddeform the links in response to the motion of theobstacles, for avoiding dynamic obstacles in dynamicenvironment.

WI.32 P0405Complex Event Processing for Object Trackingand Intrusion Detection in Wireless SensorNetworksBhargavi R, Vaidehi VIJAY KUMAR, P.T.V.Bhuvaneswari MOHAN, *P BALAMURALIDHAR, *MGirish CHANDRAAnna University, Chennai*Tata Consultancy Services

Complex Event Processing (CEP) has receivedwider acceptability due to its systematic andmultilevel architecture driven concept approach. CEPis an emerging technology in the field of dataprocessing and identifying patterns of interest frommultiple streams of events. High levels of integratedself learning applications can be developed. CEP isused in development of applications which have todeal with voluminous streams of incoming data withthe task of finding meaningful events or patterns ofevents, and respond to the events of interest in realtime. In this paper a CEP based application for objectdetection tracking in a Wireless Sensor Network(WSN) environment is proposed. Also the detection ofan intruder using semantic query processing isproposed. ESPER, an open source Complex EventProcessing engine is used to develop the application.

WI.33 P0151Multi-1D Block Matching Algorithm BasedMotion Estimation Processor using Mixed-Signal Approach

61

Shin-Yeu LIN, *Chong-Wei SU, **Jung-Shou HUANGChang Gung University*National Chiao Tung University**Elan Microelectronics Corporation

In this paper, we propose a Mixed Signal ParallelMulti 1Dimensional Block Matching Algorithm (MSPM-1D-BMA) based motion estimation (ME) processor. Incontrast to the typical 2Dimensional full search blockmatching algorithm (2DFSBMA), the MSPM-1D-BMAbased ME processor will greatly reduce the number ofdata movements in between memories. We employ avoting algorithm in the proposed ME processor toimprove the accuracy of motion vector (MV)estimation. We have demonstrated that the MVestimation accuracy achieved by the proposed MEprocessor is much better than the existing fast blockmatching algorithms and is close to 2DFSBMA. Wealso demonstrated that the speed of the proposed MEprocessor approaches two times as that of the mixedsignal 2DFSBMA based ME processor.

WI.34 P0266Vehicle Combustion Quality Monitoring: AScene Visibility-level Based Non-invasiveApproachMasood Mehmood KHANCurtin University

Pollutants interfere with light, restrict its reflectionand so impair visibility. Scene visibility level istherefore used as a measure of air quality andpollution. Treating emission efflux as “someadditional noise causing visibility impairment,” thiswork examines if the extracted visibility index from athermal infrared (TIR) image can help in qualitativeassessment of combustion efficiency. The thin-filmregime like two dimensional TIR images of unleaded-petroleum run vehicles’ exhaust-plumes were firstaccommodated for time and space relatedcompositional effects. The estimated ratios ofvisibility indices obtained from two sequential TIRimages of the same exhaust plume were comparedwith their respective electrochemically sensed levelsof oxides of nitrogen and combustibles. Initial resultssuggest that visibility indices extracted from TIRimages of emission efflux would help in distinguishinglow from high levels of emissions. TIR images cantherefore assist in qualitative assessment of enginecombustion efficiency.

WI.35 P0872Primitive-based 3D Structure Inference from aSingle 2D Image for Insect Modeling: Towardsan Electronic Field Guide for InsectIdentificationXiaozheng ZHANG, *Yongsheng GAO, Terry CAELLINational ICT Australia*Griffith University

3D insect models are useful to overcome viewingangle variations and self-occlusions in computer-assisted insect taxonomy for electronic field guides.The acquisition of 3D information is, however,unreliable due to the flexibility and small size of theinsect bodies. This paper explores how to infer 3D

insect models from a single 2D insect image, whichwill assist both insect description and identification.The 3D structure of the insect body is modeled fromtwo geometric primitives, generalized cylinders anddeformable ellipsoids. The primitives are fitted andwarped based on both edge and medial axisconstraints of the 2D image. Individualized 3Dmodels are then built to approximate the insectstructure. The proposed approach results inseemingly useful 3D insect models capable ofrepresenting the major morphological characteristicsfor a variety of insects with different body types. Thismethod could be a helpful assistance for computer-assisted insect taxonomy and insect identification byentomologists and the public.

WI.36 P0179An Approach for Raising the Accuracy of One-class ClassifiersChi-Kai WANG, Yung TING, Yi-Hung LIUChung Yuan Christian University

The support vector data description (SVDD) is amethod proposed to solve the problem of one-classclassification. It models a hypersphere around thetarget set, and by the introduction of kernelfunctions, more flexible descriptions are obtained. InSVDD, the width parameter and the penaltyparameter have to be given beforehand by the user.To automatically optimize the values for theseparameters, the error on both the target and outlierdata has to be estimated. Because no outlierexamples are available, we propose a max-min rangemethod for generating artificial outliers in this paper.By generating artificial outliers around the target set,the accuracy of classifiers will improve. At the last,we use four benchmark data sets: Iris, Wine,Balance-scale, and Ionosphere data base to validatethe approach in this research indeed has betterclassification result.

WI.37 P0372An Affine Invariant Interest Point and RegionDetector based on Gabor FiltersWanying XU, Xinsheng HUANG, Xingwei LI, YingZHANG, Jie ZHANG, Wei ZHANGNational University of Defense Technology

This paper presents a novel approach for interestpoint and region detection which is invariant to affinetransformations. Such transformations introducesignificant changes in the point location as well as inthe scale and the shape of the neighborhood of aninterest point. Our approach allows to solve for theseproblems simultaneously. The approach is based onthree key ideas: 1) Interest points can be extractedbased on local maxima of the normalized local energymaps. 2) Local extrema over scale of the normalizedenergy function indicate the presence ofcharacteristic local structures. 3) The maximumresponse along all the orientations indicates theprinciple orientation of the local structure. We firstextract interest points at multi-scales from the localenergy map constructed by Gabor filter responses,and then select points at which a local measure is

62

maximal over scales. This allows a selection ofdistinctive points for which the characteristic scale isknown. We then estimate the principle orientationthrough the orientational responses of Gabor filtersand extend the detector to affine invariance byestimating the affine shape of a point neighborhood.The characteristic scale and the affine shape ofneighborhood determine an affine invariant region foreach point. Experimental results with syntheticimages and natural images show the affine invarianceperformance of our approach. Comparativeevaluation using the repeatability criteriademonstrates the comparable performance in thepresence of large viewpoint changes.

WI.38 P0399An Efficient Illumination Normalization Methodin a Transformed DomainZhichao LIAN, Meng Joo ERNanyang Technological University

This paper proposes a novel illuminationnormalization approach with low computationcomplexity for face recognition. In this proposedmethod, a block-wise Walsh-Hadamard transform(WHT) is employed in the logarithm domain. Anappropriate number of low-frequency WHTcoefficients are zeroed to compensate for illuminationvariations. Experiments on different databasesdemonstrate that the proposed method obtainsresults comparable to those of conventional DiscreteCosine Transform method but with a higherefficiency. It also achieves better performances forcases with larger illumination variations. Furthermore,both analytical proof and experimental resultsdemonstrate that principal component analysis (PCA)can be directly implemented in the WHT domain.

WI.39 P0638A Comparative Study of Age-Invariant FaceRecognition with Different FeatureRepresentationsMeng CUI, Jiwen LU, Yap-Peng TANNanyang Technological University

Age invariant face recognition is an important yetless investigated problem in the face recognitioncommunity. In this paper, we empirically evaluatestate-of-the-art facial feature representations for age-invariant face recognition. Three representativefeatures including local binary pattern (LBP), Gaborwavelets and gradient orientation pyramid (GOP)were applied, followed by a principal componentanalysis (PCA) to reduce the dimensions of theextracted features. Experimental results on theMORPH database, one of the largest publicly availableface dataset containing thousands of longitudinalimages are presented. Experimental results show thatGabor wavelets feature with five scales and eightorientations is the optimal feature representationmethod for age-invariant face recognition.

WI.40 P0907A Weighted Voting Scheme for Recognition of

Faces with Illumination VariationAmirhosein NABATCHIAN, Esam ABDEL-RAHEEM,Majid AHMADIUniversity of Windsor

A new method for face recognition based onweighted votes on different sub-images of a picture isproposed. The proposed method is robust underillumination variations and achieves the illuminationinvariants based on the reflectance-illuminationmodel. The proposed method does not require anyprior information about the face shape or illuminationand can be applied on each image separately. It doesnot need multiple images in training stage to get theillumination invariants and is computationallyefficient. Support vector machines are used asclassifier. Several experiments are performed on YaleB and CMU-PIE databases. The system achieved99.82% recognition rate in the Yale B and 99.74%for the CMU-PIE database.

WI.41 P1063Combined Local and Holistic Facial Features forAge-DeterminationKhoa LUU, Tien Dai BUI, Ching Y. SUEN, *KarlRICANEKConcordia University*University of North Carolina Wilmington

This paper presents an advanced age-determination technique that combines holistic andlocal features derived from an image of the face. A30×1 Active Appearance Model (AAM) linearencoding of each face is produced to work as holisticfeatures. Meanwhile, local features are extracted byusing Local Ternary Patterns (LTP). These combinedfeatures are used to classify faces into one of twoage groups (age-classification). An age-determinationfunction is then constructed for each age group inaccordance with physiological growth periods forhumans - pre-adult (youth) and adult. Compared topublished results, this method yields the highestaccuracy rates in overall mean absolute error (MAE),mean absolute error per decade of life (MAE/D), andcumulative match score.

WI.42 P0667Low Level Data Fusion of Laser and MonocularColor Camera using Occupancy GridFrameworkQadeer BAIG, Olivier AYCARDUniversity of Grenoble 1

In this paper we have developed a technique forlow level data fusion between laser and monocularcolor camera using occupancy grid framework in thecontext of internal representation of externalenvironment for object detection. Based on a smallvariant of background subtraction technique weconstruct an occupancy grid for camera and fuse itwith the one constructed for laser to get a combinedview. The results obtained using Cycab simulatorprepared by INRIA show the effectiveness of ourtechnique.

63

Session TA1Date : Thursday, 09 December 2010Time : 1000 - 1200Venue : Ball Room

Robust Control IChairs : Pierre APKARIAN

ONERAHisaya FUJIOKAKyoto University

TA1.1 P0989 1000 - 1020Extended Blending Techniques withApplications in Robust Tracking Control andFault-tolerant ControlJun XU, *Lihua XIE, Kai Yew LUMNational University of Singapore*Nanyang Technological University

In this paper, we extend the blending techniqueintroduced by Blanchini et. al. to a general case.Different from the existing technique, where all thecolumns of the distribution matrices of the exogenousinputs should be linearly independent of each other,we only require that part of the columns are linearlyindependent, and the other matrices have full columnranks. Based on this blending technique, we presenta novel approach to the robust fault-tolerant controlscheme and robust tracking problem. The novelty liesin the fact that we can separately design thecontroller gains for different performance indices,while simultaneously achieving all the indices via anelegant construction of an overall controller. Thedesign procedure is based on LMI techniques andbasic linear algebra tools.

TA1.2 P0225 1020 - 1040Nonsmooth μ SynthesisPierre APKARIAN, *Hoang Duong TUANONERA*The University of New South Wales

We revisit robust complex- and mixed-musynthesis problems based on upper bounds and showthat they can be recast as specially structuredcontroller design programs. The proposedreformulations suggest a streamlined handling of musynthesis problems using recently developed (local)nonsmooth optimization methods where both scalingsor multipliers and a controller of given structure areobtained simultaneously. A first cut of the nonsmoothcode for structured H-infinity synthesis is madeavailable through the MATLAB R2010b Prerelease,Robust Control Toolbox Version 3.5 developed by TheMathWorks, Inc.

TA1.3 P0603 1040 - 1100Characterizing Approximated Differentiators inDigital ControlHisaya FUJIOKA

Kyoto UniversityCharacterizations of the approximated

differentiator used in digital control are derived. Thegap between the pure differentiator and itsapproximation is modeled as an additiveperturbation. Two types of quadraticcharacterizations are obtained. One is the L2 gainand the other is related to the passivity property. Theresults guarantee that the perturbation is small whenthe sampling period is small enough, and can be usedfor robust control design when the perturbation is notnegligible.

TA1.4 P0370 1100 - 1120Mixed Nonlinear-Sliding Mode Control of anUnmanned Farm Tractor in the Presence ofSlidingAlexey S MATVEEV, *Michael HOY, *Andrey VSAVKINSaint Petersburg University*The University of New South Wales

The paper considers the problem of automaticpath tracking by autonomous farming vehiclessubject to wheel slips, which are characteristic foragricultural applications. Two guidance laws areproposed to solve this problem, and both explicitlytake into account the constraints on the steeringangle and ensure tracking an arbitrarily curved path.The first law is implemented by the pure sliding-modecontroller, whereas the second one combines thesliding mode approach with a smooth nonlinearcontrol law, using control chattering at the reducedamplitude as compared with the first law.Mathematically rigorous proofs of global convergenceand robust stability of the proposed guidance lawsare presented. In doing so, the slipping effects aretreated as bounded uncertainties. Simulation resultsconfirm the applicability and performance of theproposed guidance approach.

TA1.5 P0323 1120 - 1140An Integral SMC Based Approach to Time-delaySystem TrackingGang LIU, *Alan ZINOBER, **Yuri SHTESSELXi'an Jiaotong-Liverpool University*The University of Sheffield**University of Alabama in Huntsville

Output tracking in SISO fully linearizable nonlinearsystems with a time delay is considered using integralsliding mode control based technique. Using Padeapproximations for the delay, the actual delayedoutput is replaced by its approximation, by which theproblem is reduced to the tracking of a nonminimumphase control system. This system is transformed intoa corresponding state tracking problem, where thestate tracking profiles are generated by the equationsof the stable system centre. The integral sliding modecontrol approach is developed and good outputtracking results are obtained. Smith Predictor is usedto compensate the difference between the actualdelayed output and its approximation and slidingmode observer/first order exact sliding mode

64

differentiator is used to deal with the perturbedsystem. A one-link robot arm example is used toshow the effectiveness of this proposed method.

TA1.6 P0368 1140 - 1200Navigation of an Unmanned Helicopter inUrban EnvironmentsMichael HOY, Andrey V SAVKIN, *Matt GARRATTThe University of New South Wales*UNSW at the Australian Defence Force Academy

When employing autonomous vehicles, it isdesirable to use controllers which can be rigorouslyshown to always ensure safety is maintained. In thismanuscript we compare two approaches for theproblem of navigation through environmentscontaining obstacles. The first uses boundaryfollowing to maintain an avoidance distance toobstacles, and the second uses a MPC-type algorithmto plan short range trajectories around detectedobstacles, while ensuring the vehicle can be broughtto a halt within the sensor radius. The controllers aresubjected to analysis for robustness, and simulationsare carried out with both a simple second order linearmodel and a realistic helicopter model for verification.The controller that planned ahead was found to givesignificantly better trajectories.

Session TA2Date : Thursday, 09 December 2010Time : 1000 - 1200Venue : Canary

Intelligent SystemsChairs : Philippe MORIGNOT

CEAInsu SONGJames Cook University Australia,Singapore campus

TA2.1 P0915 1000 - 1020Conditioned Switching between Two-Degree-of-Freedom Controllers for Plants withChanging DynamicsJoseph J YAME, Hanping QIAOUniversity Henri Poincare

This paper presents a self-conditionedimplementation of any two-degree-of-freedom(TDOF) controller which allows a conditioned transferwhen switching between a controller acting in closed-loop and another controller in a bank of idlecontrollers waiting to take over the control loop.Here, the notion of conditioned transfer is used tomean that after switching, good tracking performanceis ensured albeit a jump at plant input may occur.This notion of conditioned transfer is different fromthe notion of bumpless transfer which refers to therequirement that no bump occurs at the plant inputwhen switching between different controllers. A mainfeature of our study is that the switching amongcontrollers is driven by the changing dynamics of theplant and moreover the technique is model-free. Thisis in contrast with most works reported in theliterature on bumpless switching, where switchingoccurs between controllers driving a plant with fixeddynamics.

TA2.2 P0427 1020 - 1040Estimation of Next Human Behavior and itsTiming for Human Behavior SupportKae DOKI, *Kohjiro HASHIMOTO, *Shinji DOKI,*Shigeru OKUMA, Akihiro TORIIAichi Institute of Technology*Nagoya University

We have proposed a modeling and recognitionmethod of human behaviors in this research. In theproposed modeling method, we have assumed that aperson changes his behavior according to the changeof the situation around him, and this concept isexpressed by If-Then-Rules, which are calledbehavior rules. In behavior rules, the change of thesituation around a person is described by HiddenMarkov Model(HMM) which models multi-dimensionaltime series sensing data. In this research, a supportsystem for human driving behaviors has beenassumed as an example of application of theproposed model. In order to realize the system whichshows a driver his next behavior, we propose an

65

estimation method of the next human behavior andthe timing of its execution. The usefulness of theproposed method is examined through experimentalresults of behavior recognition with the constructedsystem.

TA2.3 P0428 1040 - 1100A Metamodel for Background Ozone Levelusing Radial Basis Function Neural NetworksHerman WAHID, *Quang P HA, **Hiep NQUYEN-DUCUniversity of Technology Sydney*University of Technology, Sydney**Climate Change and Water, NSW

In air quality modelling, determination of thebackground ozone level is essential as it highlyaffects the accuracy of the photochemical air qualitymodel. It is known that the background ozone level,especially in urban areas, has been changing over theyears. Unfortunately, the reasons of that alterationwere not clear and the background ozone itself wasnot easily derived in practice. In this paper, a newbackground ozone model will be developed by usingthe ozone ambient quality data and themeteorological data at the several stations in theSydney basin. To accomplish the modelling process,an adaptively-tuned radial basis function neuralnetwork metamodel is proposed and utilised in thesimulation. Different input parameters are consideredto evaluate their influence on the constructedbackground ozone model. The proposed model,subject to some statistical criteria, demonstrates itscapability of estimating the background ozone levelwith a reasonably good accuracy.

TA2.4 P0505 1100 - 1120Hardware-in-the-Loop Simulation of AutomaticSteering Control: Outer-Loop and Inner–LoopControl DesignPoh Ping EM, Khisbullah HUDHA, Mohd Hanif BinHARUN, *Hishamuddin JAMALUDDINUniversiti Teknikal Malaysia Melaka*Universiti Teknologi Malaysia

This paper presents a 9-degree of freedom (DOF)vehicle model combined with a closed loop drivermodel for the purpose of developing vehicle lateralcontrol. The driver model was developed to controlthe steering angle and uses the lookup table path asa reference for the control input. The proposed outer-loop controller structure for the driver model is acombination of proportional gain control with a yaweffect adaptive fuzzy logic control. A stepper motormodel, rack and pinion model, and kinematics modelof the steering system are also briefly introduced asan inner-loop sub-system for stepper motor actuatedsteering (SMAS) system. The proposed innerloopcontroller is a closed-loop positioning control for thestepper motor. The performance of the outer-loopand inner-loop controllers were evaluated usingpredefined trajectory for lanekeeping and double lanechange (DLC) maneuvers at 80 km/h constant speed.Both of the controller’s software-in-theloopsimulations (SILS) results were validated using an

instrumented automatic steering test rig through thehardwarein- the-loop simulation (HILS). The SILS andHILS results show that the proposed driver model iscapable of improving the Y-axis trajectory error andmaneuvers significantly and the proposed SMASsystem is capable of tracking the desired steeringangle position and producing the front wheel steerangle for the use of vehicle model.

TA2.5 P0978 1120 - 1140Automatic Synthesis of Reactive AgentsInsu SONG, *Guido GOVERNATORI, JoachimDIEDERICHJames Cook University Australia, Singapore campus*National ICT Australia

This paper introduces a new approach todesigning smart control chips that enables automaticsynthesis of real-time control systems from agentspecifications. An agent specification is compiled intoa hardware description format, such as RTL-VHDL(Register Transfer Level--VLSI Hardware DescriptionLanguage) or RTL Verilog, which is synthesized usingcomputer-assisted tools to develop ASIC masks orFPGA configurations. A rule-based specificationlanguage called Layered Argumentation System (LAS)is defined and a sound and complete mapping toVerilog is developed. LAS combines fuzzy reasoningand non-monotonic reasoning. This enables chipdesigners to capture commonsense knowledge andconcepts having varying degrees of confidencecollaboratively and incrementally.

TA2.6 P1054 1140 - 1200Generating Scenarios for a Mobile Robot withan Arm - Case Study: Assistance forHandicapped PersonsPhilippe MORIGNOT, Mariette SOURY, ChristopheLEROUX, Helena VOROBIEVA, Patrick HEDECEA

In this paper, we present a mobile robot with anarm and a gripper, which can generate its ownscenarios before executing them. Hand-writtenscenarios, as we previously did, are not applicableany longer, since a wider range of scenarios is aimedat, which cannot be predicted in advance. Ourapproach involves task planning, for scenariogeneration, and finite state automation, for scenarioexecution. Our robot is used for servicinghandicapped or ageing persons in their apartment.

66

Session TA3Date : Thursday, 09 December 2010Time : 1000 - 1200Venue : Kingfisher

Localization, Navigation and Mapping IIChairs : K SRIDHARAN

Indian Institute of Technology,MadrasJack Jianguo WANGUniversity of Technology, Sydney

TA3.1 P0342 1000 - 1020Maintaining Visibility of a Moving Target:Maximizing Escape Time vs. Exposure TimeIbrahim AL-BLUWI, *Ashraf ELNAGARUniversite de Toulouse*University of Sharjah

This paper presents a novel approach for theproblem of tracking a moving target in a globaldynamic environment. The robot has to move suchthat it keeps the target visible for the longest timepossible, and at the same time, avoid colliding withany of the moving obstacles. This paper presents asolution that is based on the idea of three interactingcomponents which perform: tracking, collisionavoidance and motion selection. The proposedsolution is validated using a comprehensive set ofsimulations, which show that transition from trackingin static environments to tracking in dynamicenvironments can be done without much loss in robotsafety or tracking ability.

TA3.2 P0402 1020 - 1040An Omnidirectional Time-of-Flight Camera andits Application to Indoor SLAMKatrin PIRKER, Matthias RÜTHER, Horst BISCHOF,*Gerald SCHWEIGHOFER, *Heinz MAYERGraz University of Technology*Joanneum Research GmbH

Photonic mixer devices (PMDs) are able to createreliable depth maps of indoor environments. Yet,their application in mobile robotics, especially insimultaneous localisation and mapping (SLAM)applications, is hampered by the limited field of view.Enhancing the field of view by optical devices is nottrivial, because the active light source and the sensorrays need to be redirected in a defined manner. Inthis work we propose an omnidirectional PMD sensorwhich is well suited for indoor SLAM and easy tocalibrate. Using a single sensor and multiple planarmirrors, we are able to reliably navigate in indoorenvironments to create geometrically consistentmaps, even on optically difficult surfaces.

TA3.3 P0406 1040 - 1100Ground-Plane Classification for RobotNavigation - Combining Multiple Cues Towarda Visual-Based Learning System

Tobias LOW, Antoine MANZANERAEcole Nationale Supérieure de Techniques Avancées

This paper describes a vision-based ground-planeclassification system for autonomous indoor mobile-robot that takes advantage of the synergy incombining together multiple visual-cues. A prioriknowledge of the environment is important in manybiological systems, in parallel with their reactivesystems. As such, a learning model approach is takenhere for the classification of the ground/object space,initialised through a new Distributed-Fusion (D-Fusion) method that captures colour and texturaldata using Superpixels. A Markov Random Field(MRF) network is then used to classify, regularise,employ a priori constraints, and merge additionalground/object information provided by other visualcues (such as motion) to improve classificationimages. The developed system can classify indoortest-set ground-plane surfaces with an average true-positive to false-positive rate of 90.92% to 7.78%respectively on test-set data. The system has beendesigned in mind to fuse a variety of different visual-cues. Consequently it can be customised to fitdifferent situations and/or sensory architecturesaccordingly.

TA3.4 P0677 1100 - 1120Mapping Large Scale Environments byCombining Particle Filter and InformationFilterMahesh MOHAN, K. Madhava KRISHNAInternational Institute of Information Technology,Hyderabad

This paper presents two approaches to combinetwo popular mapping strategies, namely ParticleFilters and the Information Filters. The first methoddescribes how the Particle Filter can be incorporatedinto the Information Filter framework, by buildinglocal submaps using the Particle Filter and combiningthem using an Information Filter to obtain a globalmap. Using the Particle Filter locally reduces thelinearization errors and is useful in handlingambiguous data associations, while the InformationFilter keeps track of uncertainty over long periods oftime, thereby avoiding FastSLAM's tendency tobecome overconfident. The second method showshow the Information Filter can be used in the ParticleFilter framework as a simple means of rememberingthe filter's uncertainty. This can then used torepopulate particles while closing loops. This not onlyhandles nonlinearities, but is also more robust forloop closing because, unlike the Particle Filter, theInformation Filter does not exhibit forgetfulness of atrajectory's past.

TA3.5 P1010 1120 - 1140Map-aided 6-DOF Relative Pose Estimation forMonocular SLAM using Sparse InformationFiltersZhan WANG, Gamini DISSANAYAKEUniversity of Technology, Sydney

This paper addresses the problem of mapping

67

three dimensional environments from a sequence ofimages taken by a calibrated camera, andsimultaneously generating the camera motiontrajectory. This is the Monocular SLAM problem inrobotics, and is akin to the Structure from Motion(SFM) problem in computer vision. We present anovel map-aided 6-DOF relative pose estimationmethod based on a new formulation of the MonocularSLAM that is able to provide better initial estimates ofnew camera poses than the simple triangulationtraditionally used in this context. The ’6-DOF’ meansrelative to the map which itself is up to anunobservable scale. The proposed pose estimatoralso allows more effective outlier rejection inmatching features present in the map and featuresextracted from two consecutive images. OurMonocular SLAM algorithm is able to deal witharbitrary camera motion, making the smooth motionassumption, which is required by the typically usedconstant velocity model, unnecessary. In the newMonocular SLAM formulation, the measurements ofextracted features from images are partitioned intothose used for the estimation of the environment andthose used for estimating the camera motion. Thenew formulation enables the current map estimate toaid achieving the full 6-DOF relative pose estimationup to the mapping scale while maximally exploitingthe geometry information in images. Experimentresults are provided to verify the proposed algorithm.

TA3.6 P1103 1140 - 1200Robotic Mapping with Simple Sensing andProcessing Hardware - Algorithm andArchitectureLeena VACHHANI, *K SRIDHARANIndian Institute of Technology, Bombay*Indian Institute of Technology, Madras

This paper considers the problem of generating amap of an indoor environment by a mobile robotwhen constraints are placed on sensing andprocessing hardware. In particular, we present analgorithm for Voronoi diagram generation by a mobilerobot equipped merely with inexpensive ultrasonicsensors and a low-end Field Programmable GateArray (FPGA) device. The algorithm is based on thedefinition of the Voronoi diagram in terms of theperpendicular-bisector. Deviations from the Voronoidiagram of the (reference) point on the robot arecorrected by a novel algorithm that is devoid ofdivision and floating-point operations. An efficientarchitecture and experiments with an FPGA-basedrobot are also presented.

Session TA4Date : Thursday, 09 December 2010Time : 1000 - 1200Venue : Nightingale

Instrumentation SystemsChairs : Changyun WEN

Nanyang Technological University

TA4.1 P0871 1000 - 1020A New Methodology of Continuous andNoninvasive Blood Pressure Meausrement byPulse Wave VelocityYan CHEN, Changyun WEN, *Guocai TAO, *Min BINanyang Technological University*Southwest Hospital

Blood pressure (BP) is one of the most importantphysiological parameters reflecting cardiovascularstatus of people. Continuous and non-invasive BPmeasurement, which can provide beat-to-beat BPinformation and call attention to the doctor if thepatent is in danger during surgery, plays a veryimportant role in safety control in operation andintensive care. In this paper, a novel and completemethodology of continuous and non-invasive BPmeasurement by PWV without calibration by othermethods such as cuff sphygmomanometer isproposed. Critical parts of this study include modelingBP-PWV relationship and identifying two PWVsrelative to systolic blood pressure (SBP) and diastolicblood pressure (DBP) respectively. Some benchmarkmodels for 20-year-old and 60-year-old age groupsare developed and applied to BP measurement inclinical trials. The clinically obtained results meet thetest standard in ANSI/AAMI SP10, which attests thefeasibility of the measuring method. Lastly, thecontinuous SBP and DBP measurement results byPWV method are shown.

TA4.2 P1049 1020 - 1040Mechanoinduction of Reduction in the Stiffnessof Zebrafish ChorionJoo Hoo NAM, Peter C Y CHEN, *Zhe LU, **HongLUO, Ruowen GE, **Wei LINNational University of Singapore*University of Toronto**Singapore Institute of Manufacturing Technology

In this paper, we report our investigation on theforce-induced responses of zebrafish chorion beforeand after being perturbed by an externally appliedprescribed force. Our experimental results show thatapplication of an external force leads to a reductionin the stiffness of the zebrafish chorion, and thatvariation in such a reduction is strongly influenced bythe perturbation time. Our results provide evidencesupporting the hypothesis that the stiffness of somecellular organism may be modified by the controlledapplication of an appropriate mechanical force, anddemonstrate the potential of force-feedback controlin micro/nano-manipulation as a promising

68

engineering technique for studying disease propertiesand manipulating cellular biological functions.

TA4.3 P0409 1040 - 1100Approximate Entropy-based Leak Detectionusing Artificial Neural Network in WaterDistribution PipelinesJin YANG, Yumei WEN, Ping LIChongqing University

Correlation techniques are widely used to locateleaks in buried water pipes. However, a distinct peakin the cross-correlation of two spatially separatelycollected acoustic signals may result from a non-leakacoustic source outside the pipe. And the peak notrelated to a real leak will result in a false leaklocation. So it is necessary to determine whether ornot a real leak exists beforehand. In this paper, anew leak detection method using approximateentropy is proposed to discriminate the leak acousticsignals from the non-leak signals. In this method, theautocorrelation function values for the delay largerthan the correlation length of the signal, not thesignal itself or its entire autocorrelation functionvalues, are used to extract or evaluate the self-similarity degree of the signal by the approximateentropy. A neural-network approach has beendeveloped as a classifier, which uses the identifiedself-similarity degrees as the network inputs. Theproposed leak detection method has been employedto identify the leak in the buried water pipelines, andachieved a 92.5% correct detection rate.

TA4.4 P0596 1100 - 1120Service Provider Architecture for DynamicallyReconfigurable Virtual Instruments inNetworked EnvironmentsSundaramurthy RAMALINGAM, Dananjayan PERUMALPondicherry Engineering College

This paper proposes an online service providerarchitecture for reconfigurable, expandable andcoherent virtual instruments. Using an FPGA chip withembedded software-core processor which providesboth software-programmable and hardware-reconfigurable abilities, it is possible for us toconstruct a hardware kernel that can serve as thekernel of versatile measurement instruments.Reprogramming the embedded processors andreconfiguring the internal FPGA logic in the hardwarekernel, we can implement various instrumentalfunctions. The possibility of modifying themeasurement procedure simply by changing thealgorithm executed, which imitates hardware kernelfor measurement instruments, by the computer-based architecture/software-core processor withoutreplacing the hardware components makes theexperimental activity easier. But the existingreconfigurable virtual instrument setups make theuser to reinvent the wheel and make measurementprocess a tedious one. In this paper a novel serviceprovider architecture is proposed which allows theuser to dynamically reconfigure the hardwaredepending upon their measurement needs. Withadaptively reconfiguring the hardware kernel, SDI

concept can easily respond to the rapidly changinguser-application-specified needs in measurement andtest markets in a distributed measurement system(DMS)

TA4.5 P0700 1120 - 1140Unsteady Flow Rate Generating System for Gasusing Isothermal Tank based on DisturbanceObserverXiaoxin WANG, Tao WANG, Guangzheng PENG,*Yanbing TIANBeijing Institute of Technology*Qingdao Technological University

A Generating system of unsteady flow rate for gasusing an isothermal tank is proposed. With the servovalves installed at upstream and downstream of theisothermal tank, the unsteady flow can be completelycontrolled. Moreover, by analyzing the characteristicsof the servo valve, a feed forward compensator isdesigned to improve the response of the system. Afuzzy-PID controller is introduced for the highnonlinearity of compressible fluids control system.Differential operation is involved in the controlprocess, which may amplify the noise greatly. Hencea disturbance observer is employed to reduce thenoise to a permissible range. The experiment showsthat the frequency of the unsteady flow rate obtainedfrom the generating system is up to 10 Hz.

TA4.6 P0754 1140 - 1200Prototype Development of Hybrid TemperatureRecorder Monitoring SystemOlarn WONGWIRAT, Keelati OONCHOM, BenjapornARNUTTINANONKing Mongkut's Institute of Technology Ladkrabang

Currently, the calibration of burn-in chambers usedin a HDDs (Hard Disk Drives) production line requiresan engineer to manually monitor temperatures at thehybrid temperature recorders throughout the entireperiod of calibration process. It is a time consumingoperation and considered an inefficient process, dueto a large number of burn-in chambers used in theproduction line comparing with a limited number ofengineers to operate. Therefore, this paper describesthe prototype development of hybrid temperaturerecorder monitoring system implemented to improvethe current calibration process of burn-in chamber.The work in this paper applies an object-orientedanalysis and design methodology (OOADM) in thedevelopment process. The proposed systemcomprises two subsystems, i.e., a temperaturerecording subsystem and a monitoring subsystem,interconnected through a wireless network. Thetemperature recording subsystem performs capturingtemperatures from sensors inside the burn-inchamber and sending the temperature data to acentral server at the monitoring subsystem in theengineering office for monitoring in real-time. Whenthe errors occur during the calibration process, thealarm signal is displayed and sent to the engineer foralerting. Furthermore, the temperature data is storedin a database for future verification and reference.

69

Session TA5 - Invited SessionDate : Thursday, 09 December 2010Time : 1000 - 1140Venue : Pelican

Collaborative UAV and UGV SystemsChairs : Patrick DOHERTY

Linköping University

TA5.1 P1151 1000 - 1020Cooperative Ground Target Tracking withInput ConstraintsSenqiang ZHU, Danwei WANGNanyang Technological University

This paper considers the problem of cooperativetarget tracking with multiple unmanned aerialvehicles (UAVs) subject to input constraints.Cooperation of multiple UAVs to track a movingtarget can provide better performance and enhancethe robustness. However, the physical constraints ofthe UAVs pose a significant challenge on the UAVcontroller design. In this paper, the relative courserate controller for a single UAV is firstly developedbased on a guidance vector field. Cooperative controlstrategy of multiple UAVs is studied and a variableairspeed controller is proposed to regulate temporalseparation of UAVs. Simulation results are providedto demonstrate the effectiveness of the proposedapproach.

TA5.2 P1152 1020 - 1040Formation Control of Multi-robot SystemsShuai LIU, *Chun-Lin CHEN, Lihua XIE, *Yeong-HwaCHANGNanyang Technological University*Chang Gung University

In this paper, we consider formation controlproblem for multi-robot systems under an undirectedcommunication network. All the robots will track aleader, while form a desired formation. The leadercan be static or dynamic. A distributed formationcontroller with neighbors’ input information isapplied. For practical implementation, control inputinformation from neighbors can only be received aftersome time delays. It is therefore shown that thedistributed control protocol using time delayedcontrol input information from neighbors guaranteesthe formation of the multi-agent system for anynonnegative delay. We will implement the newprotocol on the Amigo robots. The experimentalresults will demonstrate the effectiveness of the newprotocol.

TA5.3 P1154 1040 - 1100Federated DyKnow, a Distributed InformationFusion System for Collaborative UAVsFredrik HEINTZ, Patrick DOHERTYLinköping University

As unmanned aerial vehicle (UAV) applications are

becoming more complex and covering larger physicalareas there is an increasing need for multiple UAVs tocooperatively solve problems. To produce morecomplete and accurate information about theenvironment we present the DyKnow Federationframework for distributed fusion among collaborativeUAVs. A federation is created and maintained using amulti-agent delegation framework which allows high-level specification and reasoning about resourcebounded cooperative problem solving. When thefederation is set up, local information is transparentlyshared between the agents according to specification.The work is presented in the context of a multi UAVtraffic monitoring scenario.

TA5.4 P1155 1100 - 1120Generating UAV Communication Networks forMonitoring and SurveillancePer-Magnus OLSSON, Jonas KVARNSTRÖM, PatrickDOHERTY, Oleg BURDAKOV, Kaj HOLMBERGLinköping University

An important use of unmanned aerial vehicles issurveillance of distant targets, where sensorinformation must quickly be transmitted back to abase station. In many cases, high uninterruptedbandwidth requires line-of-sight between sender andtransmitter to minimize quality degradation.Communication range is typically limited, especiallywhen smaller UAVs are used. Both problems can besolved by creating relay chains for surveillance of asingle target, and relay trees for simultaneoussurveillance of multiple targets. In this paper, weshow how such chains and trees can be calculated.For relay chains we create a set of chains offeringdifferent trade-offs between the number of UAVs inthe chain and the chain’s cost. We also show newresults on how relay trees can be quickly calculatedand then incrementally improved if necessary.Encouraging empirical results for improvement ofrelay trees are presented.

TA5.5 P1156 1120 - 1140Automated Planning for Collaborative UAVSystemsJonas KVARNSTRÖM, Patrick DOHERTYLinköping University

Mission planning for collaborative UnmannedAircraft Systems (UAS:s) is a complex topic whichinvolves trade-offs between the degree ofcentralization or decentralization required, the degreeof abstraction in which plans are generated, and thedegree to which such plans are distributed amongparticipating UAS:s. In realistic environments such asthose found in natural and man-made catastropheswhere emergency services personnel are involved, acertain degree of centralization and abstraction isnecessary in order for those in charge to understandand eventually sign off on potential plans. It is alsoquite often the case that unconstrained distribution ofactions is inconsistent with the loosely coupledinteractions and dependencies which arise betweencollaborating systems. In this article, we present a

70

new planning algorithm for collaborative UAS:s basedon combining ideas from forward chaining planningwith partial-order planning leading to a new hybridpartial order forward-chaining (POFC) frameworkwhich meets the requirements on centralization,abstraction and distribution we find in realisticemergency services settings.

Session TA6Date : Thursday, 09 December 2010Time : 1000 - 1200Venue : Oriole

Image/Video Analysis IChairs : Mani Maran RATNAM

Universiti Sains MalaysiaYongqiang YENanjing University of Aeronautics andAstronautics

TA6.1 P0254 1000 - 1020Focus Measure in a Liquid-Filled Diaphragm(LFD) Lens using Passive Auto-focus MethodSoran Jalal ABDULLAH, Mani Maran RATNAM, ZahurinSAMADUniversiti Sains Malaysia

Auto-focusing in imaging systems depends on thedetermination of the correct image focus criterion. Inthis research, the image captured by a liquid-filleddiaphragm (LFD) fluid lens was analyzed todetermine a focus measure criterion that can be usedto establish the correct focus and thus quantify theimage quality. The LFD lens was actuated using astepper-motor driven syringe mechanism. The lensdiaphragm was made of polydimethylsiloxane (PDMS)polymer that exhibit good optical properties. The lensfocal length was controlled by varying the fluidvolume within the diaphragm lens. A CCD camerawas attached to the fluid lens to capture live imagesof a binary target. The edge slope width (ESW) of thepixel intensity profile across the white-to-blacktransition region in a binary target was used as thefocus measure. The experiments carried out showedthe viability of the proposed focus measure criterionfor automatically focusing the image formed by adiaphragm-type fluid lens.

TA6.2 P0281 1020 - 1040Online Background Learning for Illumination-robust Foreground DetectionDawei LI, Lihong XU, *Erik GOODMANTongji University*Michigan State University

This paper presents a background modelingalgorithm and a foreground detecting method whichis robust against illumination change, providing anovel and practical choice for intelligent videosurveillance systems using static cameras. This paperfirst introduces an online Expectation Maximizationalgorithm which is developed from the basic batchedition to update the mixture models in real time.Then a spherical K-means clustering method is usedto provide more accurate direction for the update ofGaussian Mixture Models after a deep study of RGBspace features under illumination changes.Foreground detection is carried out using a statisticalframework and RGB pixel intensity judgments. Theresults show the proposed algorithm outcompete

71

several classic methods in efficiency, accuracy, androbustness to perturbations from illuminationchanges, on a sampling of problems.

TA6.3 P0315 1040 - 1100An Automatic System for Multi-View FaceDetection and Pose EstimationYing YING, Han WANG, *Jian XUNanyang Technological University*Singapore Institute of Manufacturing Technology

Over the past few years, multi-view face detectionissue has become one of the most attractive researchtopics in the field of computer vision. In this paper, anovel automatic system for multi-view face detectionand pose estimation is proposed. Our approachadopts modified appearance-based learning methodsto build corresponding face detectors and poseestimators, and detects multi-view faces according toa coarse-to-fine structure. The experiments not onlydemonstrate the ability of our system toautomatically identify facial images with a highdegree of accuracy, but also verify its ability toestimate the pose angles.

TA6.4 P0346 1100 - 1120Nonuniformity Correction of Infrared Imagesbased on Bivariate Quadratic ModelXiubao SUI, Qian CHEN, Guohua GU, Ning LIUNanjing University of Science and Technology

The spatial fixed-pattern noise (FPN) compromisesseverely the quality of the acquired imagery, evenmakes such images inappropriate for someapplications. In order to lower the FPN, some criticalnonuniformity correction (NUC) algorithms such asNUC based on linear model, scene-based NUC and soon have been developed. But each algorithm hassome drawbacks: restricted application in smalldynamic range of objects temperature, lowperformance under the drift of the environmenttemperature and complex calculations. In thesecases, we develop a bivariate and quadratic model ofthe FPA and the NUC technique based on the model.The proposed method does not need anyassumptions and is a good solution for hardwareimplementation. It overcomes the drawbacks of thecritical algorithm mentioned above. The lastsimulations and experiments show that the proposedalgorithm exhibits a superior correction effect in bothlarge objects temperature range and environmenttemperature range.

TA6.5 P0587 1120 - 1140Preliminary Investigation of ThermalHartmann Wavefront SensingLiping ZHAO, *Kelvin Jian Aun OOI, Xiang LI, *LayKee ANGSingapore Institute of Manufacturing Technology*Nanyang Technological University

We present a novel technique for thermal imagingin capitalisation on the Hartmann wavefront sensing

technology for purposes of 3-D sensing. The issues ofreflection induced noises, diffuse nature of thermalradiation, and directionality of emissivity, which mightaffect the results of thermal Hartmann wavefrontsensing, are also discussed.

TA6.6 P0976 1140 - 1200A Novel Fractional-order Signal ProcessingBased Edge Detection MethodHaibo YANG, Yongqiang YE, *Danwei WANG, BinJIANGNanjing University of Aeronautics and Astronautics*Nanyang Technological University

Image edge detection is a classic problem ofmachine vision and image processing. Edge detectionoften uses an integer-order differential operation.The paper adopts fractional differentiation andintegration to obtain a new edge detection operator.The performances in terms of detection accuracy andnoise immunity of the new operator are comparedwith those of the traditional operators throughexamples. The comparison shows that the newoperator is promising.

72

Session TA7 - Invited SessionDate : Thursday, 09 December 2010Time : 1000 - 1200Venue : Bluebird

Computer Vision & Pattern RecognitionChairs : Soodamani RAMALINGAM

University of Hertfordshire

TA7.1 P1107 1000 - 1020License Plate Localisation based onMorphological OperationsXiaojun ZHAI, Faycal BENSAALI, SoodamaniRAMALINGAMUniversity of Hertfordshire

Automatic Number Plate Recognition (ANPR)systems allow users to track, identify and monitormoving vehicles by automatically extracting theirnumber plates. This paper presents an improvedmethod to locate car plates in an ANPR system. Theproposed method is based on morphological openand close operations where different StructuringElements (SE) are used to maximally eliminate non-plate region and enhance plate region. This methodhas been tested using a database of UK numberplates and results achieved have shown significantimprovements in terms of the detection rate compareto other existing plate localisation systems.

TA7.2 P1109 1020 - 1040Edge Detection Comparison for License PlateDetectionZuwena MUSOROMY, Soodamani RAMALINGAM,*Nico BEKOOYUniversity of Hertfordshire*CitySync Ltd.

The detection of license plate region is the mostimportant part of a vehicle’s license plate recognitionprocess followed by plate segmentation and opticalcharacter recognition. Edge detection is commonlyused in license plate detection as a preprocessingtechnique. This paper compares the performance ofthe image enhancement filters when used in edgedetection algorithms combined with connectedcomponent analysis to extract license plate region.The experimental comparison of Canny, Kirsch,Rothwell, Sobel, Laplace and SUSAN edge detectorson gray scale images shows that Canny yields highplate detection of 98.2% tested on 45,032 UK imagescontaining license plates at 720X288 resolutioncaptured under various illumination conditions. Theaverage processing time of one image is 56.4 ms.

TA7.3 P1113 1040 - 1100Efficient Character Segmentation on CarLicense PlatesLihong ZHENG, Junbin GAO, *Xiangjian HECharles Sturt University*University of Technology, Sydney

In this paper an improved hill climbing algorithmbased method is presented to cut character out ofthe license plate images. Although there are manyexisting commercial LPR systems, with poorillumination conditions and moving vehicle theaccuracy impaired. After examination and comparisonof two different types of image segmentationapproaches, the hill climbing algorithm based methodgave a better image segmentation results. The hillclimbing algorithm was modified by introducingautomatic parameter determination and smartsearching. After modification it efficiently detects thepeaks (local maxima) that represent different clustersin the global histogram of an image. The process issuccessful by getting a clean license plate imageremoving all unwanted areas. While testing by theOCR software, the experimental results show a highaccuracy of image segmentation and significantlyhigher recognition rate after non-character areas areremoved. The recognition rate increased from about30.6% before our proposed process to about 91.3%after all unwanted non-character areas are removed.Hence, the overall recognition accuracy of LPR wasimproved.

TA7.4 P0429 1100 - 1120Human Detection using Local Shape and Non-Redundant Binary PatternsDuc Thanh NGUYEN, Wanqing LI, Philip OGUNBONAUniversity of Wollongong

Motivated by the advantages of using shapematching technique in detecting objects in variouspostures and viewpoints and the discriminative powerof local patterns in object recognition, this paperproposes a human detection method combining bothshape and appearance cues. In particular, localshapes of the body parts are detected using templatematching. Based on body parts’ shapes, localappearance features are extracted. We introduce anovel local binary pattern (LBP) descriptor, calledNon-Redundant LBP (NRLBP), to encode localappearance of human. The proposed method wasevaluated and compared with other state-of-the-arthuman detection methods on two commonly useddatasets: MIT and INRIA pedestrian test sets. Wealso performed extensive experiments on selectingappropriate parameters as well as verifying theimprovement of the proposed method through allstages of the framework.

TA7.5 P0548 1120 - 1140Graph Matching Based Hand PostureRecognition using Neuro-Biologically InspiredFeaturesPramod Kumar PISHARADY, Prahlad VADAKKEPAT, AiPoh LOHNational University of Singapore

An elastic graph matching algorithm usingbiologically inspired features is proposed for therecognition of hand postures. Each node in the graphis labeled using an image feature extracted using thecomputational model of the ventral stream of visual

73

cortex. The graph nodes are assigned togeometrically significant positions in the hand image,and, the model graphs are created. Bunch graphmethod is used for modeling the variability in handposture appearance. Recognition of a hand posture isdone by the elastic graph matching between themodel graphs and the input image. A radial basisfunction is used as the similarity function for thematching process. The proposed algorithm is testedon a 10 class hand posture database which consistsof 478 grey scale images with light and darkbackgrounds. The algorithm provided betterrecognition accuracy (96.35%) compared to thereported results (93.77%) in the literature.

TA7.6 P1127 1140 - 1200Face Recognition using Discrete CosineTransform and Fisher Linear DiscriminantVaidehi VIJAY KUMAR, *Naresh Babu N. T, HAVINASH, M D VIMAL, A SUMITRA, **BalmuralidharP, **M Girish CHANDRAAnna University, Chennai*Madras Institute of Technology, Chennai**Tata Consultancy Services

In this paper, an efficient method for facerecognition based on the Discrete Cosine Transform(DCT), Fisher Linear Discriminant (FLD) and classifieris presented. First, the dimensionality of the originalface image is reduced using the DCT and illuminationvariations are alleviated by discarding the first fewlow-frequency DCT coefficients. FLD is applied to theselected DCT coefficients to discriminate the invariantfacial features. The KNN classifier is used for therecognition of the faces using the features extractedfrom the FLD. Simulation results show that theproposed system achieves better performance withhigh training and high recognition rate as well as verygood illumination robustness.

Session TM1Date : Thursday, 09 December 2010Time : 1310 - 1510Venue : Ball Room

Panel Session on “Control and Autonomy:Challenges and Opportunities”Chairs : Professor Yeng Chai SOH

Nanyang Technological University

Panelists:Professor Pedro ALBERTOSPolytechnic University of Valencia, SpainProfessor Patrick DOHERTYLinkoping University, SwedenProfessor Minyue FUThe University of Newcastle, AustraliaProfessor Daniela RUSMIT, USAProfessor Bruno SICILIANOUniversity of Naples Federico II, Italy

74

Session TM2Date : Thursday, 09 December 2010Time : 1310 - 1510Venue : Canary

Feature Extraction, Grouping andSegmentation IChairs : Kezhi MAO

Nanyang Technological UniversityAndrzej SLUZEKNanyang Technological University

TM2.1 P0415 1310 - 1330Detection and Segmentation of Near-duplicateFragments in Random ImagesAndrzej SLUZEK, *Mariusz PARADOWSKI, DuanduanYANGNanyang Technological University*Wroclaw University of Technology

Retrieval of near-duplicate image fragments is oneof the most challenging problems is CBIR (content-based image retrieval). The objective is to identifyalmost the same fragments in random images ofunpredictable contents. Such fragments usuallyrepresent identical object, though captured from adifferent viewpoint, under different photometricconditions and/or by a different camera. The paperpresents techniques developed for such applications.In general, the proposed methods are based onstatistical properties of keypoint similarities betweencompared images. In the first approach, we assumethat near-duplicates are (approximately) related byaffine transformations, i.e. the underlying objects arelocally planar. In the second approach, a wider rangeof shape distortions is acceptable. Implementations(including online detection in real-time videos) arepresented and their performances discussed.Additionally, an algorithm for a highly accuratesegmentation of detected near-duplicate fragments ispresented.

TM2.2 P0619 1330 - 1350A Systematic Approach for Rapid 3DReconstruction from PhotosetsHoai Nam LE TRAN, Sornum Kabilen KADIRVELEN,Hock Soon SEAH, Wentong CAI, Malcolm Yoke HeanLOW, Suiping ZHOU, Michael Harold LEESNanyang Technological University

We present an interactive framework for rapid 3Dreconstruction of objects from sets of photographs.Our framework implements advanced 2D image-based modeling techniques that accurately computegeometries of digital photo-realistic 3D models in arapid and robust method. It is implemented in anapplication tool to assist users to perform featuredetection, matching and meshing. Our systemdemonstrates that with a few user’s interventionthrough our application tool, efficient generation of3D object models can be generated from a set ofphotos.

TM2.3 P0713 1350 - 1410Image Segmentation of UV Pattern forAutomatic Paper-Money InspectionKeon-Ho LEE, Tae-Hyoung PARKChungbuk National University

We propose an image segmentation method forautomatic paper-money inspection system. The UV(ultra violet) patterns embedded in the paper moneyshould be segmented to determine whether themoney is genuine or not. We acquired the image bydirectional lighting system using UV LEDs. In order tosegment the pattern from the background, we applythe Gaussian mixture model to consider the multi-modal characteristics of the histogram. Experimentalresults are presented to verify the performance of theproposed method.

TM2.4 P0731 1410 - 1430Fruit Detection, Tracking, and 3DReconstruction for Crop Mapping and YieldEstimationJednipat MOONRINTA, SupawadeeCHAIVIVATRAKUL, Matthew N DAILEY, MongkolEKPANYAPONGAsian Institute of Technology

Through automated agricultural inspection,farmers can potentially achieve better productivityand accurately predict yields and crop quality. Avariety of sensors can be used for agriculturalinspection, but the cheapest and most information-rich is the video camera. We collect data in the fieldfrom a monocular camera fixed to a mobile inspectionplatform. For purposes of pineapple crop mappingand yield prediction, we propose an image processingframework for in-field fruit detection, tracking, and3D reconstruction. We perform a series ofexperiments on feature point extraction using Harris,SIFT, and SURF features, feature point descriptionusing SIFT and SURF descriptors, feature pointclassification using SVMs, fruit region tracking usingblob tracking, and 3D reconstruction using structurefrom motion and robust ellipsoid estimationtechniques. We find that SURF feature points anddescriptors provide the best tradeoff betweenprocessing time and classification accuracy and thatthe method is sufficiently accurate for fruit regiondetection. Our preliminary results for fruit regiontracking and 3D fruit reconstruction are promising.We plan further work towards development of auseful aid to help farmers manage their farms.

TM2.5 P0755 1430 - 1450Applying Training Hidden Features to JointCurve Evolution for Brain MRI SegmentationMahshid FARZINFAR, Eam Khwang TEOH, *ZhongXUENanyang Technological University*Weill Medical College of Cornell University

According to the level of information provided inimages, segmentation techniques can be categorized

75

into two groups. One is region-labeling, which obeysthe intensity-based classification methods. Althoughmodeling the tissue intensity is straightforward byapplying local statistical methods and spatialdependencies, the results might suffer from noiseand incomplete data. The second group of techniquesapplies active contour models, in which the objectiveis to find the optimal partition of the image domainusing a closed or open curve by using priorconstraints on the shape variation. However,estimating optimal curve is intractable due to theincomplete observation data. This paper extends apreviously reported joint active contour model formedical image segmentation in a new Expectation-Maximization (EM) framework, wherein the evolutioncurve is constrained not only by a shape-basedstatistical model but also by applying a hiddenvariable model from the image observation. In thisapproach, the hidden variable model is defined by thelocal voxel labeling computed from its likelihoodfunction, depended on the image functions and theprior anatomical knowledge. Comparative results onsegmenting putamen and caudate shapes in MR brainimages confirmed both robustness and accuracy ofthe proposed curve evolution algorithm.

TM2.6 P0847 1450 - 1510Designing Compact Gabor Filter Banks forEfficient Texture Feature ExtractionWeitao LI, *Kezhi MAO, **Hong ZHANG, TianyouCHAINortheastern University*Nanyang Technological University**University of Alberta

Texture feature has been widely used in imagesegmentation, classification, retrieval and manyothers. Among various approaches to texture featureextraction, Gabor filtering has emerged as one of themost popular in recent years. Gabor filter-basedtexture feature extractor is in fact a Gabor filter bankdefined by its parameters including frequencies,orientations and smoothing parameters of theGaussian envelope. In the literature, theseparameters are often set by trial and error, based onthe experience of the user, and the Gabor filter banksthus designed are often over-sized. To address theproblem mentioned above, we propose to designcompact Gabor filter banks by incorporating filterselection in this study. We develop a newMahalanobis separability measure-based supervisedapproach to address the need of texture featureextraction. The strengths of our methods are twofold.Firstly, the proposed method provides a systematicway for Gabor filter bank design to avoid man-madebias. Secondly, the compact filter banks thusdesigned overcomes the problem of redundant orinsignificant/irrelevant filter banks, and this in turnleads to improved performance of textureclassification. Experimental results on benchmarkdatasets demonstrate the effectiveness of ourproposed approach.

Session TM6Date : Thursday, 09 December 2010Time : 1310 - 1510Venue : Oriole

Computational Intelligence in VisionChairs : Eric SUNG

Nanyang Technological University

TM6.1 P0351 1310 - 1330Profit of Extending Standard RelationalDatabases with the Intelligent Cluster Index(ICIx)Sebastian LEUOTH, Alexander ADAM, *WolfgangBENNDimensio Informatics GmbH*Chemnitz University of Technology

In this paper, we present a strategy to reduce theprocessing time needed for selection operations withmany attributes in standard database systems. Theseproblems mostly occur in data mining, data analysis,information retrieval, and applications with highcombinatorial complexity. In these systems, standardindexes do not gain a satisfying performance.Currently, this problem is tackled using morecomputing power or special solutions instead ofstandard databases. Our approach is to interpret thequeries as high-dimensional point or range queries.Thus, we provide a “real” solution to answeringcomplex queries rather than merely postponing theproblems using technical methods. We show thebenefit of multi-dimensional data structures. Thisbenefit can be transferred to a lot of applications(e.g. Business Intelligence, Bill Of MaterialsExplosion, DNA Sequence Search), not only advancedapplications of database systems like GIS, CAD, ormultimedia. Finally, a very small sample data set isused to show the profit of our approach and wepresent possible integration points of the ICIx intostandard relational database.

TM6.2 P0495 1330 - 1350Ensembles of Novel Visual KeywordsDescriptors for Image CategorizationAzizi ABDULLAH, *Remco C VELTKAMP, **Marco AWIERINGUniversiti Kebangsaan Malaysia*Utrecht University**University of Groningen

Object recognition systems need effective imagedescriptors to obtain good performance levels.Currently, the most widely used image descriptor isthe SIFT descriptor that computes histograms oforientation gradients around points in an image. Apossible problem of this approach is that the numberof features becomes very large when a dense grid isused where the histograms are computed andcombined for many different points. The currentdominating solution to this problem is to use aclustering method to create a visual codebook that is

76

exploited by an appearance based descriptor tocreate a histogram of visual keywords present in animage. In this paper we introduce several novel bagof visual keywords methods and compare them withthe currently dominating hard bag-of-features(HBOF) approach that uses a hard assignmentscheme to compute cluster frequencies. Furthermore,we combine all descriptors with a spatial pyramid andtwo ensemble classifiers. Experimental results on 10and 101 classes of the Caltech-101 object databaseshow that our novel methods significantly outperformthe traditional HBOF approach and that our ensemblemethods obtain state-of-the-art performance levels.

TM6.3 P1045 1350 - 1410Real-Time Dense Digital Elevation MapEstimation using Laserscanner and CameraSLAM ProcessFlorent MALARTRE, *Pierre DELMAS, RolandCHAPUIS, *Chistophe DEBAINBlaise Pascal University*Cemagref

This paper is about environment perception fornavigation system in outdoor applications. Unlikeother approaches that try to detect an obstacle ofbinary state, we consider here a Digital Elevation Map(DEM). This map has to be built in regards to theguidance system's needs. These needs depend onthe vehicle capabilities, its dynamics constraints, itsspeed etc... Starting with the navigation system'sneeds, our goal is to estimate a precise and denseDEM. Our approach is based on a SLAM algorithmcombining a 2D rangefinder and a camera. Thus wecan estimate both the displacement between twolaser scan and have a good density of thereconstructed DEM. The approach has been validatedboth using simulated realistic data and real data inreal-time on our robot in outdoor environment(approx. 40ms per loop).

TM6.4 P1056 1410 - 1430A New SOM-based Active Contour Model usingConscience and Archiving MechanismsFereshteh SADEGHI, Hamid IZADINIA, RezaSAFABAKHSHAmirkabir University of Technology, Tehran

Active contour models are widely used inextracting object boundaries. However, most of thesemethods usually fail to capture concave boundariesproperly and impose high computational cost. In thispaper, a new SOM-based active contour model whichintroduces the Conscience and Archiving mechanisms(CASOM) is proposed to extend the Batch SOMmethod and eliminate its deficiencies. Theperformance of the proposed method is evaluated bysome experiments on a set of grayscale images.Experimental results are compared with those of theBSOM in terms of accuracy and convergence speed.The results reveal that compared to BSOM, theproposed method requires less computations forconverging to the object boundaries and extracts theboundaries of complex objects more accurately, even

in the presence of weak or broken edges.

TM6.5 P0540 1430 - 1450Color Rank and Census Transforms usingPerceptual Color ContrastGuangming XIONG, Xin LI, Jianwei GONG, HuiyanCHEN, *Dah-Jye LEEBeijing Institute of Technology*Brigham Young University

Rank and census transforms provide highresistance to radiometric distortion, vignette, andnoise because they are based on the relative orderingof local pixel intensity values rather than the pixelvalues themselves. These transforms are widely usedin many computer vision applications. An importantstep of computing these transforms is to compare orrank two grayscale values, which is very much likemeasuring color difference in color image. Colordifference between two color points at any part of auniform color space corresponds to the perceptualdifference between the two colors by the humanvision system. Based on this idea, we propose to useperceptual color contrast to implement color rank andcensus transforms and achieve this withoutsignificantly increasing the amount of data to processand without complicated computations. Furthermore,we demonstrate the feasibility of using these newtransforms to find correspondences for stereo vision.

TM6.6 P0895 1450 - 1510Stereo for Robots: Quantitative Evaluation ofEfficient and Low-memory Dense StereoAlgorithmsFederico TOMBARI, Stefano MATTOCCIA, Luigi DISTEFANOUniversity of Bologna

Despite the significant number of stereo visionalgorithms proposed in literature in the last decade,most proposals are notably computationallydemanding and/or memory hungry so that it isunfeasible to employ them in application scenariosrequiring real-time or near real-time processing onplatforms with limited resources such as embeddeddevices. In this paper, we have selected the subset ofproposals that appears more suited to the aboverequirements and, since literature lacks a propercomparison between these methods, we propose aquantitative experimental evaluation aimed athighlighting the best performing approach under thetwo criteria of accuracy and efficiency. The evaluationis performed on a standard benchmark dataset aswell as on a novel dataset, acquired by means of anactive technique, characterized by realistic workingconditions.

77

Session TP1 - Invited SessionDate : Thursday, 09 December 2010Time : 1600 - 1720Venue : Ball Room

Sensor NetworksChairs : Han-Lim CHOI

Korea Advanced Institute of Scienceand TechnologyWendong XIAOInstitute for Infocomm Research

TP1.1 P0364 1600 - 1620Consensusability of Discrete-time Multi-agentSystems via Relative Output FeedbackKeyou YOU, Lihua XIENanyang Technological University

This paper investigates the joint effects of agentdynamic and network topology on theconsensusability of linear discrete-time multi-agentsystems via relative output feedback. An observer-based distributed control protocol is proposed. Anecessary and sufficient condition forconsensusability under this control protocol is given,which explicitly reveals how the intrinsic entropy rateof the agent dynamic and the eigenratio of theundirected communication graph affectconsensusability. As a special case, the discrete-timedouble integrator system is discussed where a simplecontrol protocol directly using the two-step relativeposition feedback is provided to reach a consensus.The theoretic results are illustrated by a simulationexample.

TP1.2 P0799 1620 - 1640Analysis of Mutual Information for InformativeForecasting using Mobile SensorsHan-Lim CHOI, *Jonathan P. HOWKorea Advanced Institute of Science and Technology*Massachusetts Institute of Technology

This paper presents several analysis of mutualinformation that is often used to define the objectivefunction for trajectory planning (and scheduling) ofsensor networks, when the goal is to improve theforecast accuracy of some quantities of interest. Theapproach extends the present author’s prior work inorder to consider more general notion of verificationentities and to enable more robust decision withpotential uncertainty in the mission specifications.The expression of mutual information for windowedforecasting, in which the verification entities aredefined by a finite time window instead of a singletime instance, is derived and quantified withoutadding significant computational cost. It is alsodemonstrated that the sensitivity of mutualinformation to the variation of verification time canbe calculated in the same process of computing themutual information. Simple numerical examples arepresented for preliminary validation of theapplicability of the proposed analysis.

TP1.3 P0811 1640 - 1700Scalable Fusion with Mixture Distributions inSensor NetworksKuochu CHANG, Wei SUNGeorge Mason University

Mixture distributions such as Gaussian mixturemodel (GMM) have been used in many applicationsfor dynamic state estimation. These applicationsinclude robotics, image and acoustic processing,distributed tracking, and multisensor data fusion.However, the recursive processing of the mixturedistributions incurs rapidly growing computationalrequirements. In particular, the number ofcomponents in the mixture distribution growsexponentially when multiple of them are combined.In order to keep the computational complexitytractable, it is necessary to approximate a mixturedistribution by a reduced one with fewercomponents. Mixture reduction is traditionally doneby iteratively removing insignificantly components ormerging similar ones. However, a systematicprocedure is needed in order to ensure scalabilitywhile trading-off performance. In this paper, wepropose a recursive mixture reduction algorithm forGaussian mixture distribution with a given errorbound. To meet the error bound, we applied aconstraint optimized weight adaptation to minimizethe integrated squared error (ISE) between thereduced distribution and the original one. Withextensive simulations, we showed that the proposedalgorithm provides an efficient and effective mixturereduction performance in distributed fusionapplications.

TP1.4 P1117 1700 - 1720Framework of WSN Based Human CentricCyber Physical In-pipe Water MonitoringSystemAdnan NASIR, Boon-Hee SOONG, SelvakumaranRAMACHANDRANNanyang Technological University

We present a human centric cyber physicalframework architecture of our in-pipe watermonitoring and feedback system. This systemcomprises of the physical water distributioninfrastructure, together with the hardware andsoftware supported intelligent agents for waterallotment, leak detection and contamination spreadcontrol. An agent-based approach for connecting thecyber and physical layers is selected, where theagents get information from sensors monitoring thephysical components and provide this information tothe cyber system. The Hidden Markov Models (HMM)are briefly discussed to determine water consumptionpatterns and make decisions for cyber and physicalsystems.

78

Session TP2 - Invited SessionDate : Thursday, 09 December 2010Time : 1600 - 1740Venue : Canary

Advances in Biometric Theory andApplications IIIChairs : Kar-Ann TOH

Yonsei UniversityWei Yun YAUInstitute for Infocomm Research

TP2.1 P0798 1600 - 1620A Projection Framework for Biometric ScoresFusionKar-Ann TOHYonsei University

This paper presents a projection framework forbiometric scores fusion. Essentially, the frameworkconsists of a projection stage and a learning stage.Apart from investigating into several relatively newprojection models for biometric fusion, the projectionstage attempts to unify these models into a singleparametric structure. Three learning methods areinvestigated in conjunction with six projection modelsfor their impacts on verification accuracy expressed interms of equal error rate. An extensive experiment ofthese model and learning combinations on 32 fusiondata sets are performed in the evaluation.

TP2.2 P1132 1620 - 1640Design and Implementation of a ContactlessPalm Print and Palm Vein SensorKah Ong Michael GOH, Connie TEE, *Andrew BengJin TEOHMultimedia University*Yonsei University

This paper presents an innovative contactless palmprint and palm vein recognition system. We design ahand sensor that could capture the palm print andpalm vein image using low-resolution web camera.Both the visible and infrared images can be capturedat the same time, and we do not need specializedinfrared sensor to image the vein pattern. The designof the device is simple and low-cost. The subject canbe shielded completely from the complication ofundergoing two separate acquiring processes. Weallow subjects to position their hands freely abovethe sensor and they can move their hands during theacquisition process. In order to obtain clear image ofthe palm vascular pattern, we propose a novel imageenhancement technique called local-ridge-enhancement (LRE). The proposed method removesillumination error while keeping good contrastbetween the print/vein pattern and the backgroundimage. Besides, we introduce a simple yet robustdirectional coding technique to encode the palm printand palm vein features in bit string representation.The bit string representation offers speedy templatematching and enables more effective template

storage and retrieval. The scores output by the palmprint and palm vein experts are fused using SupportVector Machine. The fusion of these features yieldspromising result for practical implementation.

TP2.3 P1149 1640 - 1700A Metric of Identification Performance ofBiometrics based on Information ContentSeira HIDANO, Tetsushi OHKI, Naohisa KOMATSU,*Kenta TAKAHASHIWaseda University*Hitachi Ltd.

We propose the minimum distance entropy (MDE)as a metric of biometric information content. TheMDE is the probability that two biometric samplescorrespond exactly expressed in information contentand can be calculated through the experiment forinterpersonal matching using a set of biometricsamples. This metric makes it possible for certainbiometrics not only to be compared with otherbiometrics but also to be partially compared withpersonal authentication using passwords, PIN, orother methods in regard to the identificationperformance or the security. In this paper, wediscuss the metric in terms of information theory andshow how to evaluate it. Then, as an example, weapply it to a fingerprint system and evaluatefingerprint information content through simulations.

TP2.4 P0719 1700 - 1720A Comparative Study of Local FeatureExtraction for Age EstimationSung Eun CHOI, Youn Joo LEE, Sung Joo LEE, *KangRyoung PARK, Jaihie KIMYonsei University*Dongguk University

Many age estimation methods have been proposedfor various applications such as Age Specific HumanComputer Interaction (ASHCI) system, age simulationsystem and so on. Because the performance of theage estimation is greatly affected by the agingfeature, the aging feature extraction from facialimages is very important. The aging features used inprevious works can be divided into global and localfeatures. As global features, Active AppearanceModels (AAM) was mainly used for age estimation inprevious works. However, AAM is not enough torepresent local features such as wrinkle and skin.Therefore, the research about local features isrequired. In previous works, local features weregenerally used to determine age group rather thandetailed age, and the comparative studies aboutvarious local features extraction methods were notconducted. In this paper, the performances of sobelfilter, difference image between original andsmoothed image, ideal high pass filter (IHPF),gaussian high pass filter (GHPF), Haar andDaubechies discrete wavelet transform (DWT) arecompared for extracting local features and detailedage estimation is performed by Support VectorRegression (SVR) on BERC and PAL aging database.The experimental results show that local features can

79

be used for detailed age estimation and GHPF gives abetter performance than other methods.

TP2.5 P0726 1720 - 1740Dynamic Detection Rate-based Bit Allocationfor Biometric DiscretizationMeng-Hui LIM, Beng Jin Andrew TEOHYonsei University

Biometric discretization converts extractedbiometric features into a binary string via a processof segmenting every one-dimensional feature spaceinto possibly distinct multiple intervals and encodingevery interval-captured feature elementcorrespondingly. Eventually, the individual binaryoutput of every feature element is concatenated intoa binary string. To the best of our knowledge,Detection Rate Optimized Bit Allocation (DROBA)scheme is currently the most effective biometricquantization scheme, offering its capability inassigning bits dynamically for each user-specificfeature. However, we discover that DROBA suffersfrom potential discriminative feature miss-detectionsand under-quantized conditions. This paper highlightssuch drawbacks and improves upon DROBA byincorporating a dynamic searching method toefficiently recapture such miss-detected features.Experimental results illustrating significantimprovements in classification accuracy justify thepracticality of our approach.

Session TP3Date : Thursday, 09 December 2010Time : 1600 - 1800Venue : Kingfisher

Robust Control IIChairs : Janardhanan

SIVARAMAKRISHNANIndian Institute of Technology, DelhiKrzysztof J LATAWIECOpole University of Technology

TP3.1 P0380 1600 - 1620Reaching Law Based Sliding Mode Control forDiscrete MIMO SystemsMija S J, Susy THOMASNational Institute of Technology Calicut

This paper analyzes the effectiveness of thereaching law approach for sliding mode control ofdiscrete time systems. The method is extended forsliding mode control for multi-input multi-output(MIMO) systems. Modified reaching law for reducingthe chattering is proposed. The performance of theproposed controller is analyzed by carrying outsimulation studies on MIMO systems. Theperformance of the controller is also compared withthat of the equivalent control.

TP3.2 P0386 1620 - 1640Design of Multirate Output Sampling basedLinear Functional Observer and State FeedbackJanardhanan SIVARAMAKRISHNAN, NeeliSATYANARAYANAIndian Institute of Technology, Delhi

This paper addresses the problem of realizing apre-defined linear function of a state using multiratesampling for the purpose of implementing statefeedback control law. For that, the functional shouldbe expressed as a linear combination of output vectorconstructed by sampling system output at faster rate.This method can also be applied for systems havingunobservable modes. The proposed method has theadvantage that the number of output samplesrequired can be less than the observability index.Numerical example is considered to illustrate theprocedure of proposed method.

TP3.3 P1020 1640 - 1700Right/Left Inverses of Polynomial Matrices forControl, Communications and VisionWojciech P HUNEK, Krzysztof J LATAWIEC, MarianLUKANISZYNOpole University of Technology

This paper reviews various right/left inverses ofpolynomial matrices, introduced by the authors as aresult of their research works on minimum variancecontrol for LTI MIMO systems. In consequence, anew, general, Smith-factorization based inverse is

80

presented. The problem of selection of stableinverses, in particular pole-free inverses, is alsotackled. Applications in control, communications andvision, in particular concerning perfect signaltransmission/recovery, are indicated.

TP3.4 P0935 1700 - 1720Passivity-Preserving Frequency WeightedModel Order Reduction Techniques for GeneralLarge-Scale RLC SystemsWan Mariam WAN MUDA, Victor SREERAM, HerbertHo-Ching IUUniversity of Western Australia

This paper presents passivity-preservingfrequency-weighted model order reduction (MOR)techniques for general large-scale RLC systems.Three techniques (Enns', Wang's and Lin and Chiu's)which preserve only stability and not passivity aregeneralized to include passivity in this paper.Conditions under which the passivity is preserved arealso derived. Simulation results are also presented toshow the effectiveness of the proposedgeneralization.

TP3.5 P0673 1720 - 1740Automotive Driveline Control by a NonlinearNonparametric QFT MethodAhmed Fahmy ABASS, Andrew Thomas SHENTONUniversity of Liverpool

A new nonlinear Quantitative Feedback Theory(QFT) methodology is presented and applied to thecontrol of the automotive driveline with nonlinearclutch and backlash. A driveline model is presentedwhich incorporates the nonlinear clutchcharacteristics and the coupling backlash realisticallysandwiched by compliant inertias. The model givesresponses that are representative of typical drivelinesystems, with the backlash accentuating themagnitude of oscillatory shuffle response. ANonparametric (NP) identification method is given fordirectly developing the QFT templates of the LinearTime-Invariant Equivalent (LTIE) plant model set.The identification uses a Discrete Fourier Transform(DFT) to convert the system data to the frequencydomain (FD). To mitigate against windowing errorsthe frequency response (FR) data is locally smoothedby a sliding frequency weighting function. The LTIEset is obtained as set of NP identified modelsobtained directly from typical vehicle experimentalinput-output test data. Non-minimum phase (NMP)QFT bounds are determined from the nominal NPplant by determining the Hilbert transform of theNMP plant. The controller design methodology isapplied to the model and it is shown that predefinedrobust tracking and disturbance responseperformance specifications can be systematicallyobtained, without subsequent heuristic parametertuning.

TP3.6 P1058 1740 - 1800A Sligthly Improved Robust Adaptive Observer

for Actuator Fault DetectionKamel MENIGHED, Joseph J YAME, ChristopheAUBRUNUniversity Henri Poincare

In this paper, we address the problem of actuatorfault detection and estimation for a plant in thepresence of external disturbances. We design arobust fault estimation technique based on anadaptive observer with enhanced speed ofadaptation. The construction of the observer iscarried out through a transformation of the plantmodel into its special coordinate basis (SCB) form.This transformation allows a decoupling of the faultestimates from the disturbances. Thanks to thisdecoupling, the standard adaptive observer is slightlymodified to achieve faster fault estimation inpresence of disturbances and plant-modeluncertainties. Simulation experiments are presentedto illustrate the effectiveness of the proposedadaptive scheme.

81

Session TP4 - Invited SessionDate : Thursday, 09 December 2010Time : 1600 - 1800Venue : Nightingale

Intelligent Diagnosis and PrognosisChairs : Junhong ZHOU

Singapore Institute of ManufacturingTechnology

TP4.1 P0923 1600 - 1620An exTS based Neuro-Fuzzy Algorithm forPrognostics and Tool Condition MonitoringOlivier MASSOL, *Xiang LI, Rafael GOURIVEAU,*Junhong ZHOU, *Oon Peen GANUniversité de Franche Comte*Singapore Institute of Manufacturing Technology

The growing interest in predictive maintenancemakes industrials and researchers turning themselvesto artificial intelligence methods for fulfilling the tasksof condition monitoring and prognostics. Within thisframe, the general purpose of this paper is toinvestigate the capabilities of an Evolving eXtendedTakagi Sugeno (exTS) based neuro-fuzzy algorithm topredict the tool condition in high-speed machiningconditions. The performance of evolving Neuro-Fuzzymodel is compared with an Adaptive Neuro-FuzzyInference System (ANFIS) and a Multiple RegressionModel (MRM) in term of accuracy and reliabilitythrough a case study of tool condition monitoring.The reliability of exTS also investigated.

TP4.2 P0924 1620 - 1640A Knowledge Base System for RotaryEquipment Fault Detection and DiagnosisJunhong ZHOU, *Yen Tat, Louis WEE, *Zhao WeiZHONGSingapore Institute of Manufacturing Technology*Nanyang Technological University

This paper studies the fault detection anddiagnosis for the most common faults in the rotaryequipment. Large amount of experiments are carriedout on the machinery fault simulator for simulatingdifferent types of rotary machine faults. The studycovers from different type of data acquisition sensors,different signal processing and feature extractiontechniques. A hierarchical rule-based fault detectionsystem which comprises of a knowledge base coupledwith an inference engine is proposed. Theknowledge-base that maps the fault mode to signalprocessing and detection methods is built up. Therule-based fault detection system capable of assistingmechanics and engineers to deal with fault diagnosisof the rotary equipment is presented.

TP4.3 P0962 1640 - 1700FDI and Fault Estimation based on DifferentialEvolution and Analytical Redundancy RelationsMing YU, Danwei WANG, *Ming LUO, *Danhong

ZHANGNanyang Technological University*Singapore Institute of Manufacturing Technology

This article studies fault detection and isolation(FDI) and fault estimation in complex hybrid systems.The FDI approach is based on a set of unifiedconstraints, called augmented Global AnalyticalRedundancy Relations (AGARRs), to detect andisolate the faults. In order to estimate the magnitudeof the fault parameter, the fault candidates, adifferential evolution (DE) method is employed. Thisdeveloped method is applicable to estimation ofmultiple faults of parametric and nonparametricnature. Simulation is carried out to verify theeffectiveness of the proposed method in a frontsteering system of a CyCab mobile robot withmultiple faults.

TP4.4 P0977 1700 - 1720Continuous Health Assessment using a SingleHidden Markov ModelOmid GERAMIFARD, Jianxin XU, *Junhong ZHOU,*Xiang LINational University of Singapore*Singapore Institute of Manufacturing Technology

In this paper, two temporal models, HiddenMarkov Model and Auto Regressive Moving Averagemodel with exogenous inputs (ARMAX), are used forhealth condition monitoring of the cutter in a millingmachine. Dataset is acquired through real time forcesignal sensing. A heuristic statistical approach is usedto select dominant features, leading to the selectionof 3 dominant features from the 16-dimensionalfeature space. Subsequently Hidden Markov Modeland ARMAX model have been trained to predict thewearing status of the cutter in the milling machine.Suitability of these approaches are investigated andcompared.

TP4.5 P1115 1720 - 1740Towards Machine Diagnostics on ChipShen-Hoong ONG, Kiah-Mok GOH, Hian-Leng CHAN,*Teck-Yian LIM, *Keck Voon LINGSingapore Institute of Manufacturing Technology*Nanyang Technological University

Failures of critical factory equipment are one ofthe main reasons for production stoppages and aregular maintenance schedule is required to ensurecontinuous operation of production line. However,regular maintenance can be both costly andinefficient. Vibrations are present in all machinerywith moving parts and it had always been regardedas an indicator of the health and condition of rotatingmachinery. With advances in technology, what usedto be a mechanic's hunch can now be measured andwith reasonable accuracy. This paper present aproof-of-concept implementation of machinediagnostic on chip using an intelligent wireless sensornode, capable of processing and analyzing vibrationsignals from Micro-electromechanical (MEMS)sensors. Special design considerations were alsomade to allow the wireless sensor node to be

82

reconfigurable and modular at both the hardware andsoftware level. A custom fixed-point library wasimplemented to tackle accuracy issues in dealing withsmall numbers without the use of floating pointnumbers in resource scarce platforms.

TP4.6 P1128 1740 - 1800Flute based Analysis of Ball-Nose MillingSignals using Continuous Wavelet AnalysisFeaturesAmin TORABI JAHROMI, *Olivier MASSOL, Meng JooER, **Xiang LI, **Beng Siong LIM, **Oon Peen GAN,**Sheng HUANG, et alNanyang Technological University*Université de Franche Comte**Singapore Institute of Manufacturing Technology

Surface Finishing and End Milling are among themost sophisticated manufacturing processes. For theindustry to improve the quality of it end-lineproducts, it is important for improving thperformance of these processes by having adescriptive reference mode. Using this referencemodel, non-intrusive prediction of the resultingsurface quality and the tool status can be accuratelyconducted. Many modeling techniques have beenused in literature. Since there are no report ofsuccess on a general model that support all the toolspecifications, cutting conditions and correlation ofthe too-health, cutting signals and resulting surfaceroughness or tool-wear, the researches based on theseverable available AI techniques and differentsensor signals and there features are ongoing. Thispaper investigates the existing correlation betweenthe resulted wavelet coefficients and ball-nose tool-wear using Cascaded Feed Forward Neural Networks(CFFNN). Considering the changes in the shape ofthe signals during the cutting process and thesimilarity of the resulting signals during the cuttingprocess and the similarity of the resulting signals tosome mother wavelets and the lack of literature onwavelet analysis for ball-nose cutters’ signals, thisspecific analysis is chosen. CFFNN is also selected forits capability to deal with a non-linear process andbeing comparatively simple. The results aresatisfying with the proposed structure. More studiesfor the optimal structure and features are expected inthe future.

Session TP5 - Invited SessionDate : Thursday, 09 December 2010Time : 1600 - 1800Venue : Pelican

Variable Structure Control SystemsChairs : Suiyang KHOO

Deakin University

TP5.1 P1089 1600 - 1620Natural Logarithm Sliding Mode Control (ln-SMC) using EMRAN for Active Engine MountingSystemAndika Aji WIJAYA, Wahyudi MARTONO, RiniAKMELIAWATI, Fadly Jashi DARSIVANInternational Islamic University Malaysia

Robustness of the controller becomes majorimportant factor in active engine mounting system toensure the vibration of the engine can be attenuatedin entire range of operating speed. As robustness isthe main advantage of sliding mode control, it hasbeen proposed for vibration control applications.However, most of the proposed sliding mode controlare using linear sliding surface. In addition, there isno schematic method to determine the optimumsliding function. In this paper, natural logarithm-based sliding function is introduced. By its nature,natural logarithm function is nonlinear function. Sinceit consists of one parameter only, which is related tomaximum allowable vibration level, it is more easilyto be determined by the designer. Moreover, theadvantage of fast on-line training using ExtendedMinimum Resource Allocating Network (EMRAN)algorithm, made it possible to design feedbackcontrol law without having to obtain the precisesystem model which is normally becomes majordifficulties of sliding mode approach.

TP5.2 P1099 1620 - 1640Convergence Accuracy Analysis of DiscretizedSliding Mode Control SystemsBin WANG, Xinghuo YU, Liuping WANGRMIT University

In this paper, the convergence accuracy of systemstates in discretized sliding mode control (SMC)systems is thoroughly analyzed using big O notation.It confirms that for a n-th order system, systemstates x1, x2, ..., xn have different convergencerates.

TP5.3 P1118 1640 - 1700Navigation of Four-Wheel-Steering MobileRobots using Robust Fault-Tolerant SlidingMode ControlMadushanka Nishan DHARMAWEERA, *SuiyangKHOO, **Zhihong MANMonash University, Sunway Campus*Deakin University**Swinburne University of Technology

83

This paper addresses the leader-follower trackingproblem of a four-wheel-steering robot subjected tononlinear uncertainties. Two control laws have beendeveloped, based on the adaptive sliding modemethod and the adaptive input-output feedbacklinearization method. The proposed control schemeshave been tested by means of simulations.

TP5.4 P1145 1700 - 1720Leader-follower Consensus Control of a Classof Nonholonomic SystemsSuiyang KHOO, *Lihua XIE, **Zhihong MANDeakin University*Nanyang Technological University**Swinburne University of Technology

In this paper, we propose a systematic solution tothe leader-follower consensus of a class ofnonholonomic chained form systems. The controldesign, which is reminiscent of terminal sliding modeand multi-surface sliding mode control methods, ispresented to guarantee the convergence of multiplesliding surfaces, which also implies the convergenceof the proposed consensus error function. On thesesliding surfaces, the desired leader-followerconsensus can be reached for multi-agent networkformed by the nonholonomic chained form systems.

TP5.5 P1047 1720 - 1740Quantization of Optimal Control System &Synthesis of Optimal Controls of QuantumDiffusion ProcessBalai Chandra ROYCalcutta University

When applied quantization to classical controlsystem, quantum mechanics opens completely newperspectives; by exploiting quantum mechanicalfeatures, such as, time evolution of quantum stateand costate (generalized momenta) of theHamiltonian of the system, one can formulate andsolve the optimal control problems of the quantummechanical system more efficiently than the classicalsystem. The physical perspective of quantization ofthe optimal feedback system is outlined by exploitingthe formulation of the optimization problem of thesystem within the context of Hamilton-Jacobi theoryand Pontryagin maximum/minimum principle inabstract Hilbert space. One of the most strikingexample is the optimal control problem ofsynthesizing the feedback law of quantum diffusionprocess. The optimal diffusion process is shown tosatisfy the Schrödinger wave equation of energy withthe feedback law as potential function.

TP5.6 P0263 1740 - 1800A Nonlinear Exponential Observer for a BatchDistillationAmiya K JANAIndian Institute of Technology, Kharagpur

In this contribution, a state observer, namelynonlinear exponential observer (NEO) is proposed for

a batch distillation process. This estimation schememainly computes the imprecisely known parameter(augmented state) based on the availablemeasurement. For the representative distillation unit,the state predictor is formulated on the basis of onlya component material balance equation around thecondenser-reflux drum system. It clearly indicates theexistence of a process/model mismatch and theeffect of this discrepancy is efficiently taken care ofby the corrector part of the NEO estimator. Severalsimulation experiments have been executed to showthe observer error convergence ability. The lesscomputational requirements and simple design makethe closed-loop observer attractive for online use.

84

Session TP6 - Invited SessionDate : Thursday, 09 December 2010Time : 1600 - 1800Venue : Oriole

Autonomy of Vehicular Systems IChairs : Christian LAUGIER

INRIAPhilippe MARTINETClermont Universit´e

TP6.1 P0502 1600 - 1620A Real-Time Robust Global Localization forAutonomous Mobile Robots in LargeEnvironmentsJianping XIE, *Fawzi NASHASHIBI, Michel PARENT,**Olivier GARCIA FAVROTINRIA*Mines ParisTech**INDUCT Company

Global localization aims to estimate a robot’s posein a learned map without any prior knowledge of itsinitial pose. Achieving highly accurate globallocalization remains a challenge for autonomousmobile robots especially in large-scale unstructuredoutdoor environments. This paper introduces a real-time reliable global localization approach with thecapability of addressing the kidnapped robot problemusing only laser sensors. Our approach includes foursteps: 1) local Simultaneous Localization andMapping 2) map matching 3) position tracking and 4)localization quality evaluation. For sensor perception,we use occupancy grid method to represent robotenvironment. A novel pyramid grid-map basedcoarse-to-fine matching approach is proposed toimprove the localization accuracy. Experimentalresults including an outdoor environment of 25,000m² are presented to validate the feasibility andreliability of the proposed approach.

TP6.2 P0511 1620 - 1640The ArosDyn Project: Robust Analysis ofDynamic ScenesIgor E PAROMTCHIK, Christian LAUGIER, MathiasPERROLLAZ, Yong MAO, Amaury NEGRE,*Christopher TAYINRIA*ProBayes

The ArosDyn project aims to develop embeddedsoftware for robust analysis of dynamic scenes inurban traffic environments, in order to estimate andpredict collision risks during car driving. The on-boardtelemetric sensors (lidars) and visual sensors (stereocamera) are used to monitor the environment aroundthe car. The algorithms make use of Bayesian fusionof heterogenous sensor data. The key objective is toprocess sensor data for robust detection and trackingof multiple moving objects for estimating andpredicting collision risks in real time, in order to helpavoid potentially dangerous situations.

TP6.3 P0655 1640 - 1700Detection of Unfocused Raindrops on aWindscreen using Low Level Image ProcessingFawzi NASHASHIBI, Raoul DE CHARETTE, AlexandreLIAMines ParisTech

In a scene, rain produces a complex set of visualeffects. Obviously, such effects may infer failures inoutdoor vision-based systems which could haveimportant side-effects in terms of securityapplications. For the sake of these applications, raindetection would be useful to adjust their reliability. Inthis paper, we introduce the problem (almostunprecedented) of unfocused raindrops. Then, wepresent a first approach to detect these unfocusedraindrops on a transparent screen using a spatio-temporal approach to achieve detection in real-time.We successfully tested our algorithm for IntelligentTransport System (ITS) using an on-board cameraand thus, detected the raindrops on the windscreen.Our algorithm differs from the others in that we donot need the focus to be set on the windscreen.Therefore, it means that our algorithm may run onthe same camera sensor as the other vision-basedalgorithms.

TP6.4 P0684 1700 - 1720Visibility Distance Estimation based onStructure from MotionClément BOUSSARD, *Nicolas HAUTI`ERE, **BrigitteD'ANDRÉA-NOVELINRIA*LCPC**Mines ParisTech

It is obvious to say that perception is necessary todrive. Futhermore we can say that a good visibility isa guaranty for passengers security. The driver willadapt the vehicle speed to the offered visibility.Strong visibility reductions (dense fog for instance)are conditions of risk of accident. We thereforepropose here a method to perform an onboardestimation of the visibility distance. Once thisestimate is obtained, assistance could be offered tothe driver (eg if it runs at a speed not adapted to thecurrent visibility) or to the road infrastructuremanagement so that it can inform other users of riskon his road network. This method uses imagesacquired by an onboard camera filming the scene andthe estimation of vehicle motion. Thus, from thisinformation we will explain how we can achieve aspatial partial structure reconstruction to estimate thevisibility distance.

TP6.5 P0770 1720 - 1740Collaborative Multi-Vehicle Localization andMapping in High Clutter EnvironmentsMoratuwage M.D.P., Wijerupage Sardha WIJESOMA,Bharath KALYAN, *Nicholas M PATRIKALAKIS,Peyman MOGHADAMNanyang Technological University

85

*Massachusetts Institute of TechnologyAmong today’s robotics applications, exploration

missions in dynamic, high clutter and uncertainenvironmental conditions is quite common.Autonomous multi-vehicle systems come in handy forsuch exploration missions since a team ofautonomous vehicels can explore an environmentmore efficiently and reliably than a singleautonomous vehicle (AV). In order to improve thenavigation accuracy, especially in the absence of apriori feature maps, various simultaneous localizationand mapping (SLAM) algorithms are widely used insuch applications. As for multi-vehicle scenarios,collaborative multi-vehicle simultaneous localizationand mapping algorithm (CSLAM) is an effectivestrategy. However use of multiple AVs posesadditional scaling problems such as inter-vehicle mapfusion, and data association which needs to beaddressed. Although existing CSLAM algorithms areshown to perform quite adequately in simulations,their performance is much less to be desired in highclutter scenarios that is inevitable in actualenvironments. In this paper, we present an approachto improve the performance of a CSLAM algorithm inthe presence of high clutter, by combining aneffective clutter filter framework based on RandomFinite Sets (RFS). The performance of the improvedCSLAM algorithm is evaluated using simulationsunder varying clutter conditions.

TP6.6 P0520 1740 - 1800Hardware and Software Architecture for AUVbased on Low-cost SensorsAlain Sebastian MARTINEZ LAGUARDIA, YidierRODRIGUEZ PEREZ DE ALEJO, Luis HERNANDEZSANTANA, Carlos GUERRA MORALES, *Hichem SAHLIUniversidad Central de las Villas*Vrije Universiteit Brussel

The use of Autonomous Underwater Vehicles(AUV) as robots for exploration and oceanologyscience has been a field of interest of severaluniversities and research center's around the world inthe last decade. Cuba being a country surrounded bythe Caribbean Sea, having most of it's resources in it.Researchers from the Central University of Las Villas(UCLV) and the Hydrographic Research Center (HRC)have joined forces in the development of the HRC-AUV project, whit the objective of the implementationof a AUV capable to operate in the Cuban sea forexploration and supervision. The platform is based onlow cost sensors and hardware. In the presentdocument the hardware and software architectures ofthe AUV are presented. Also a set of experimentalresults are presented to validate the system.

Session TP7Date : Thursday, 09 December 2010Time : 1600 - 1800Venue : Bluebird

Visual ServoingChairs : Wen-Chung CHANG

National Taipei University ofTechnologyJorge POMARESUniversity of Alicante

TP7.1 P0396 1600 - 1620Direct Visual Servo Control of a Robot to TrackTrajectories in Supervision TasksGabriel J GARCIA, Carlos A JARA, Jorge POMARES,Fernando TORRESUniversity of Alicante

This paper describes the modelling, simulation andvisual servo control of a robot-arm designed forsupervision of complex task environments. This robot(mini-robot) has 2 degrees of freedom (hereafterd.o.f.) and it is attached at the end of anotheranthropomorphic robot with 7 d.o.f. (main robot). Inorder to achieve a suitable supervision, the mini-robot has to perform a given trajectory around theworkspace where the task is developed. With thepurpose of obtaining an accurate and fastsupervision, this paper presents a novel trackingmethod carried out in the image space using directvisual servo control.

TP7.2 P0507 1620 - 1640Visual Servoing using Triangulation with anOmnidirectional Multi-Camera SystemBernhard WEBER, Kolja KÜHNLENZTechnische Universität München

In this paper a new approach for robot controlusing position based visual servoing (PBVS) with anomnidirectional multi-camera system is presented.PBVS requires the explicit calculation of the positionand orientation of the robot tool. Given only imageswithout depth information, either additionalgeometric properties of the observed scene or stereocorrespondences have to be provided for poseestimation. In this paper we use triangulation ofimages given by many cameras pointing in variousdirections. We confirm this approach with somesimulational results.

TP7.3 P0560 1640 - 1700Vision-Based Path Planning with ObstacleAvoidance for Mobile Robots using LinearMatrix InequalitiesWeifeng HUANG, *Anan OSOTHSILP, FarzadPOURBOGHRATSouthern Illinois University, Carbondale*Assumption University

86

In this paper, a vision-based obstacle avoidingpath generation problem is considered forautonomous mobile robots under a top-viewworkspace. The collision-free path planning problemis converted to a convex optimization problem thatcan be solved numerically using linear matrixinequalities (LMI). A new optimal (shortest) path costformulation is given for LMI optimization using anovel Line of Sight Meshing (LSM) method. Ascompared to the traditional meshing algorithms suchas Voronoi and Delaunay, the LSM generates fewermesh cells which results in reduced computationtime. A virtual path diagram (VPD), consisting of allcollision-free piecewise straight-line paths, is thenfound for robot navigation. The LMI optimizationmethod is then used to find the optimal (shortest)collision-free path from the VPD set. In addition todistance, other constraints, such as terrain conditionand curvature of the path can be incorporated in thecost function for constrained optimal solution. Finally,a rapid prototyping (RP) environment has beendeveloped, which is used for hardwareimplementation of the algorithm. The generatedpaths are then converted to motion commands,which are sent wirelessly from a base-station tomobile robots for motion control.

TP7.4 P0611 1700 - 1720An Intelligent Space for Mobile RobotNavigation with On-Line Calibrated VisionSensorsWen-Chung CHANG, Ping-Rung CHUNational Taipei University of Technology

This paper presents a seemingly novel approach toaccomplishing mobile robot localization andnavigation in a large unknown workspace with a setof vision sensors based on an effective on-linecalibration strategy. Each adjacent IP camera pair forvisual sensing is assumed to have an overlappingfield of view, but the positions as well as orientationswith respect to either the Cartesian space or themobile robot are unknown. The idea is to control themobile robot to actively perform calibration with anystationary IP camera that can observe the five pre-selected color-coded features onboard the mobilerobot, and thus establish coordinate transformationsamong these IP cameras. In particular, thecoordinate transformation from the mobile robot tothe IP camera is recursived updated based on allobserved data and the kinematic model of the mobilerobot. The proposed system only requires a PC and anetwork of stationary IP cameras without performingoff-line calibration. Therefore, the proposed approachappears to be low-cost, effective, efficient, andflexible.

TP7.5 P0688 1720 - 1740A Novel Position-Based Visual ServoingApproach for Robust Global Stability withFeature Points Kept within the Field-of-ViewIn-Joong HA, Do-Hwan PARK, Jeong-Hun KWONSeoul National University

In this paper, we shed new light on PBVS(Position-Based Visual Servoing) by introducing theconcept of a 3D (Dimension) visible set for PBVS,which can serve to play exactly the role of the 2Dvisible set, image plane, for IBVS. Our 3D visible setis convex even in the presence of uncertainties inintrinsic and extrinsic parameters of the camera, justas the image plane is a convex set. Then, we canensure the FOV (Field of View) constraint simply bycontrolling the camera pose to follow a smoothstraight-line trajectory in our 3D visible set for PBVS,which connects two points determined by the initialand desired poses of the camera. To demonstratefurther the validity and practicality of our VS scheme,we also present various experimental results using a6 degree-of-freedom robotic manipulator.

TP7.6 P0913 1740 - 1800A Framework of Networked Visual ServoControl System with Distributed ComputationHaiyan WU, Lei LOU, Chih-Chung CHEN, SandraHIRCHE, Kolja KÜHNLENZTechnische Universität München

In this paper, a networked visual servo controlsystem with distributed computation is proposed toovercome the low sampling rate problem in vision-based control systems. A real-time image datatransmission protocol based on Realtime TransportProtocol (RTP) is developed. The captured images aresent to different processing nodes connected over acommunication network and processed in parallel.Thus, a high sampling rate of the visual feedback isachieved under a cloud image processingarchitecture. The varying image processing delaycaused by the varying number of extracted featuresand the random transmission delay are modeled as arandom process with Bernoulli distribution. By usingthe input-delay approach, the resulted networkedvisual servo control system is reformulated into astochastic continuous-time system with time-varyingdelay. Experiments on two 1-DoF linear motormodules are carried out to validate the proposedapproach. A visual servo control system withoutparallel distributed computation is implemented forcomparison. The experimental results demonstratesignificant performance improvement by theproposed approach.

87

Session TIDate : Thursday, 09 December 2010Time : 0930 - 1800Venue : Foyer

Interactive Session II

TI.1 P0247Simple Analytic Formulas for PID TuningWuhua HU, Gaoxi XIAO, Wen-Jian CAINanyang Technological University

This paper proposes simple analytic formulas forproportional-integral-derivative (PID) controllertuning for typical process models. The formulas areobtained in a similar way to the simple internal modelcontrol (SIMC) tuning rules, while the leading analysisis more delicate. Compared to SIMC counterparts, thenew tuning formulas lead to better load disturbancerejection while giving similar setpoint response andpeak sensitivity.

TI.2 P0389An Optimal Point-wise Control Method forParabolic Distributed Parameter SystemsQian LI, Ning LI, Shao-yuan LIShanghai Jiaotong University

An optimal point-wise control method for parabolicdistributed parameter systems is proposed to solvethe problem of determining both the locations ofpoint-wise controllers and the control which shouldbe exerted on each controller. For a given number ofpoint-wise controllers, the optimal point-wise controlform is given by solving a quadratic cost controlproblem, and the controller locations and the controlare determined by minimizing the quadratic controlcost performance index. The result indicates themethod proposed is effective at solving the optimalpoint-wise control problem for parabolic distributedparameter systems.

TI.3 P0453Decoupling the Control Variables of a Plant inPresence of Time DelayJulian FUCHS, Ulrich KONIGORSKITechnische Universität Darmstadt

In this contribution a concept is presented, whichoffers independent controllability of pressure andflow-rate whitin process plants. The dynamic of thoseplants is highly nonlinear and the control variablesare coupled. Furthermore, time delays have to beconsidered. The proposed control concept consists ofa nonlinear decoupling controller in combination withan extended smith predictor. The functionallity of thecontroller is demonstrated by measurements on atest bench.

TI.4 P0572Modelling of a Hot Water Drum and Heat

Exchanger Process Control Training SystemBelinda Chiew Meng CHONG, *Mohd Nor MOHDTHANUniversity Teknology MARA Malaysia*University of Tun Hussein Onn

The purpose of this paper is to study amathematical formulation of a water level andproduct liquid temperature control loop in a boilerdrum and heat exchanger process control trainingsystem. Mathematical modeling was carried out tounderstand the dynamics of the liquid producttemperature for different settings of PID controllermode. The modeling objective is to address theprediction of the plant performance with sufficientaccuracy so that experimental data can be verifiedwith the simulation results. The experimental datawas obtained from the mini plant and the simulationresults by model developed by using Simulink toolboxin Matlab software. The experimental and simulationresults that were verified exhibited fair agreement.

TI.5 P0669Fractional−Order Generalized PredictiveControl: Formulation and some PropertiesMiguel ROMERO, Angel P DE MADRID, CarolinaMANOSO, *Blas M VINAGREUniversidad Nacional de Educacion a Distancia*University of Extremadura

This paper deals with the formal formulation ofFractional−Order Generalized Predictive Control, ageneralization of GPC that makes use of fractional–order operators in its cost function. A thoroughmathematical description of FGPC is given and themain similarities and differences between GPC andFGPC are highlighted.

TI.6 P1037Harmonic Excitation Based Autotuning -Autotuning Usable for Control of BiomassCombustion ProcessesStanislav VRÁNA, Bohumil ŠULCCzech Technical University in Prague

Controlling combustion processes is an importanttask, mainly with respect to energy efficiency andecological impacts. This problem has becomesignificant with the increased use of biomass-firedboilers, including low-power boilers. The combustionprocess is a process that we have limited informationabout. The equations for the chemical reactions areknown, but in reality combustion of biomass is muchmore complicated and more variable than amathematical model can express. It is therefore notsimple to design a controller that is able to controlthe combustion process effectively in the wholerange. Small-scale biomass-fired boilers are usuallyequipped with simple two-state (on/off) controllers.Many laboratories have attempted to develop amodel of the combustion process suitable forcontroller setting and parameter adaptation, but theresults are not yet satisfactory or generally valid.Thus, advanced model-based control strategies are

88

not usable, while the use of a model-free controller isnot excluded if well-designed tuning rules can beprovided. An approach which tries to satisfy most ofthe requirements of industrial practice is presented inthis paper, which is based on an experimentallyperformed evaluation of small amplitude excitedfrequency responses with the aim to achieverecommended values of one or more control qualityindicator known e.g. from the course of the Nyquistplot. The indicators can be evaluated experimentallyin control loops involving nonlinearities. In this sense,the method has a philosophy similar to that of thepopular Ziegler and Nichols method, but nointerruption of the control process is necessary andthe amplitude of the excited oscillation can be set byoperator. The main advantages of the methodpresented here are: there is no use of anymathematical model, it can be used as an addition tothe existing controller purely by software, and thereis no need to interrupt the control process whileretuning the controller.

TI.7 P0709On-line Identification and Control Methods forPID ControllersUtkal MEHTA, Somanath MAJHIIndian Institute of Technology, Guwahati

The aim of this paper is to present a method foron-line tuning of PID controller for stable processes.Without breaking closed-loop control, an explicitprocess model is developed from a single relay testand then controller gains are re-tuned non-iterativelyto improve the performance. An explicit tuning rulesare derived with special emphasis on minimizing thecontrol efforts. Examples are given to illustrate thesimplicity and superiority of the proposed methodcompared with some existing ones.

TI.8 P0541Shaping Nonlinear Modal Manifold of InternalResonance SystemTa-ming SHIHMing Dao University

For system of internal resonance(IR) properties,there will be modal coupling effect while the systemlinear frequency is commensurable. In order todiscuss the dynamic behavior on the IR manifold, weneed to include all the coupling manifolds as oneinvariant manifold, by using the modal invarianceproperty, the system energy can be constrained inthis manifold. In this research, an extended normalform method is developed to study the nonlinearInternal Resonance normal mode. The method isbased on the traditional normal form method innonlinear dynamics analysis. By using the modalinvariance property, we will form a set of homologicalequations; and the coordinate transformation processis converted to a coefficients-to-be-determinedproblem. The proposed method gives a clear view ofthe dynamic structures on each modal manifold,which is useful for analyzing some modal couplingphenomenal such as internal resonance (IR) or modal

localization problem.

TI.9 P0565Analysis and Application of Trajectory inDiscrete-time SystemJianyin FANG, Kai LIUHenan Institute of Engineering

In this work, we focus on the properties oftrajectory in discrete-time system. under someconditions, the behavior of this class of systems'trajectory is studied. Using this properties, severalcharacterizations of discrete-time system are alsogiven, which is very helpful to study the ISS propertyof discrete-time system.

TI.10 P0639Two Flat Normal Forms for a Class of NonlinearDynamical SystemsBououden SORAYA, Boutat DRISS, *AbdessemedFOUDILPRISME-ENSI de Bourges*University of Al Jouf

In this paper we presents two new 0-flat normalforms. It deals with sufficient geometrical conditionswhich enable us to conclude if a given nonlinearcontrollable dynamical system can be transformed, bymeans of change of coordinates, to one of thesenormal forms. In the same way it gives an algorithmto compute the flat outputs. As an illustration to theproposed approach, a trajectory tracking of a planarmanipulator robot is simulated. The results obtainedare very satisfactory proving the applicability of themethod.

TI.11 P1068Nonlinear Disturbance Decoupling for aNonholonomic Mobile Robotic ManipulationPlatformJoel JIMENEZ-LOZANO, Bill GOODWINEUniversity of Notre Dame

A mobile manipulator is at the present time awidespread term to refer to robot systems built froma robotic manipulator arm mounted on a mobileplatform. A mobile manipulation system offers a dualadvantage of mobility offered by a mobile platformand dexterity offered by the manipulator. In thiswork, the tracking and nonlinear disturbancedecoupling problems are studied. We show that thissystem posses the necessary geometric structure forcomplete disturbance decoupling between theoutputs and disturbances. Simulation results obtainedfor the mobile manipulator show good performancein the presence of significant disturbances using thedesigned nonlinear controller.

TI.12 P0419Real-time INS/GPS Data Fusion using HybridAdaptive Network Based Fuzzy InferenceMalleswaran M, Mary Anita J, Sabreen SN, *Vaidehi

89

VIJAY KUMARAnna University, Tirunelveli*Anna University, Chennai

Kalman filter is used to integrate the data frominertial navigation systems (INS) and global positionsystems (GPS) for vehicle navigation to provideposition, velocity and attitude. However severaldrawbacks of Kalman filter such as noise, modeling ofthe system, initial noise parameter are restricted itsapplication and implementation. In this paper the realtime implementation of GPS/INS integration using ahybrid Adaptive Neuro Fuzzy Inference System(ANFIS) is proposed. The proposed system is trainedduring the availability of GPS signal to map the errorbetween the GPS and the INS. Then it will be used topredict the error of the INS position componentsduring GPS signal blockage. The data from GPS andINS are used to build a structured knowledge baseconsisting of behavior of the INS in some specialscenarios of vehicle motion. In the absence of theGPS information, the system will perform its task onlywith the data from INS and with the trained dataprovided by ANFIS. The system is evaluated whileconsidering several intentionally introduced GPSoutages for periods of 50 seconds, with the positionaccuracy mostly below 0.2m. The simulated resultsshow the advantages of the proposed method ofANFIS techniques for INS/GPS integration.

TI.13 P0517A Benchmarking Tool for MAV Visual PoseEstimationGim Hee LEE, Markus ACHTELIK, FriedrichFRAUNDORFER, Marc POLLEFEYS, Roland SIEGWARTETH Zurich

The large collections of datasets for researchersworking on the Simultaneous Localization andMapping problem are mostly collected from sensorssuch as wheel encoders and laser range findersmounted on ground robots. The recent growinginterest in doing visual pose estimation with camerasmounted on micro-aerial vehicles however has madethese datasets less useful. In this paper, we describeour work in creating new datasets collected from asensor suite mounted on a quadrotor platform. Oursensor suite includes a forward looking camera, adownward looking camera, an inertial measurementunit and a Vicon system for groundtruth. We proposethe use our datasets as benchmarking tools for futureworks on visual pose estimation for micro-aerialvehicles. We also show examples of how the datasetscould be used for benchmarking visual poseestimation algorithms.

TI.14 P0610Mathematical Formulation of RFID Tag Floorbased Localization and Performance Analysisfor Tag PlacementYoungsu PARK, Je Won LEE, Daehyun KIM, Jae JinJEONG, Sang-woo KIMPohang University of Science and Technology

A radio-frequency identification (RFID) tag floor

based localization is recently proposed indoor mobilerobot localization method that utilizes super-distributed RFID tag infrastructure (SDRI) installed ona working area. An RFID tag floor localization (RTFL)method is easy to scale up the working area and thenumber of robots and is reliable in the positionestimation. There have been several researches forpractical applications of the RTFL, however, theinvestigation on performance and properties of thelocalization method is still insufficient. This paperpropose a mathematical formulations of the RTFL andits performance index based on an RFID positionestimation error variance. This paper also presentssimulation results of the RTFL performances andanalysis. These results can be used for optimalinstallation of the RFID tag floor.

TI.15 P0682Ad-hoc Swarm Robotics Optimization in GridBased NavigationSiddarth JAIN, Manish SAWLANI, Vijay KumarCHANDWANIMedi-Caps Institute of Technology and Management

A technology to identify a correct remote objectand to carry it back to a base camp has become animportant utility form, in the military for identifyingimprovised explosive devices (IEDs) and for spacescientists to collect samples from mars. Multi-robotsystems can accomplish tasks that would beimpossible for a single robot to achieve. Developingmulti-agent systems with self interested agents witha large behavioural repertoire is a great challenge. Inthis paper, we combine collaborative agent behaviourto deal with these vital issues. An ad-hoc designapproach for swarm robotics optimization in gridbased navigation is presented. Implementation of lowpower CC2500 wireless communication chip is doneto accomplish multi-robot communication. Primitivebehaviors called real-time grid search, move-to-goal,avoid-static-obstacle, avoid-robot behaviors areintroduced for the path planning of multi-robot. Thesystem is based on Real-time Grid Searching andCollision free Path Tracing & Retracing algorithm. Arandomly placed remote object in the grid isidentified and is carried to the base station with theeffective coordination of the multi-robot system.

TI.16 P0690Map Based Indoor Robot Navigation andLocalization using Laser Range FinderShung Han CHO, Sangjin HONGStony Brook University

This paper presents a map based robot navigationand localization algorithm using laser range finder forindoor environments. A navigation path is given bythe sequence of grids and global map information isrepresented by the list of vertexes. The grid basednavigation facilities path planning as well ascomplements localization with priori information ofthe grid sequence. The pattern of vertexes isrepresented by distance, adjacency, and slant amongthem for the comparison between map information

90

and range data. A mobile robot is globally localizedby finding the matched pattern between the set ofvertexes from the map and the set of vertexes fromthe range data. The proposed method is verified withactual range data from laser range finder.

TI.17 P0782Improved Genetic Algorithms Based OptimumPath Planning for Mobile RobotChin Yun SOH, *Ganapathy VELAPPA, Ooi Chong LIMMonash University, Sunway Campus*University of Malaya

Improved genetic algorithms incorporate othertechniques, methods or algorithms to optimize theperformance of genetic algorithm. In this paper,improved genetic algorithms of optimum pathplanning for mobile robot navigation are proposed.An Obstacle Avoidance Algorithm (OAA) and aDistinguish Algorithm (DA) are introduced togenerate the initial population in order to improve thepath planning efficiency to select only the feasiblepaths during the evolution of genetic algorithm.Domain heuristic knowledge based crossover,mutation, refinement and deletion operators arespecifically designed to fit path planning for mobilerobots. Proposed genetic algorithms feature unique,simple path representations, and simple but effectiveevaluation methods. Simulation studies and real timeimplementations are carried out to verify and validatethe effectiveness of the proposed algorithms.

TI.18 P0851Attitude Determination of Large Non-cooperative Spacecrafts in Final ApproachXuehai GAO, Bin LIANG, Wenfu XUHarbin Institute of Technology

Due to failure of mechanisms to deploy, somelarge communication satellites lost their ability andresulted in huge economic cost. A space roboticsystem is expected to perform the on-orbit repairingmission. It is a tremendous challenge to navigate aspace robot in final approach since the targets aregenerally non-cooperative. Rectangle features, whichare common in the configuration of a satellite, can bechosen as the recognized objects. However, thesecharacters are very large. Limited by the FOV (field ofview), a monocular camera can not supply enoughinformation of the rectangles. In this paper, amethod based on monocular camera is proposed todetermine the attitude of a large non-cooperativetarget using a partial rectangle. Firstly, therelationship of a rectangle and circular points bycamera is derived and fused. Secondly, the attitudemeasurement algorithm is acquired from theconstraint and a virtual rectangle is reconstructed.Lastly, the algorithm is verified by mathematicsimulations which are very close to reality. Theresults show the validity and flexibility of theproposed method.

TI.19 P0413Modelling and Simulation of a Multi-fingeredRobotic Hand for Grasping TasksJuan Antonio CORRALES, Carlos A JARA, FernandoTORRESUniversity of Alicante

This paper develops the kinematic, dynamic andcontact models of a three-fingered robotic hand(BarrettHand) in order to obtain a completedescription of the system which is required formanipulation tasks. These models do not only takeinto account the mechanical coupling and thebreakaway mechanism of the under-actuated robotichand but they also obtain the force transmission fromthe hand to objects, which are represented astriangle meshes. The developed models have beenimplemented on a software simulator based on theEasy Java Simulations platform. Several experimentshave been performed in order to verify the accuracyof the proposed models with regard to the real physicsystem.

TI.20 P0512New Hand Posture Classification Strategy forFinding Kinematically-Feasible PrecisionGraspsPeerapong THONNAGITH, Attawith SUDSANGChulalongkorn University

We present new strategy for finding kinematically-feasible precision grasps for particular robot hand. Bycollecting samples of hand posture and classifyingthem into multi level-of-detail of posture informationfor each hand configuration, we can efficiently searchfor valid hand postures given a set of contact pointswithin reasonable processing time. The approach forcreating hand posture database can be also appliedto any kind of robot hands as long as the kinematicspecification of those hands are provided.

TI.21 P0943Study on the Structure and Control of aDexterous HandHao LIU, Tao WANG, Wei FAN, Tong ZHAOBeijing Institute of Technology

A new dexterous hand is designed and produced.The shape and structure of the dexterous hand isdescribed, focused on analyzing the layout of tendonsand the optimization of the joint radius. Furthermore,a single finger is controlled by Fuzz-PID with modelcompensation and the good control precision isachieved. Through the study on the single fingercontrol, the master-slave control of the dextroushand is realized simulating the movement of humanhand with a cyber-glove.

TI.22 P0537Coordinated Control of the PowerCompensation Equipments Located in LocalSubstations of the District Unit Power SystemSang-Deok LEE, Ji-Ho PARK, Young-Sik BAEK

91

Kyungpook National UniversityDecreasing stability of power system by voltage

instability makes that all power system face withserious problems of stability. That is attributable tothe difficulty of controlling the power system. So it isnecessary to keep the voltage of load bus in range ofthe limited level to supply high quality of electricpower. Therefore we describe and suggest on thispaper that a plan for coordination control betweenUPFC(Unified Power Flow Controller), Shuntelements(Shunt capacitor & Shunt reactor) andULTC(Under Load Tap Changer) amongcompensators and also describe the method to keepor control the voltage of power system in allowableranges.

TI.23 P0306Local Image Feature Matching for ObjectRecognitionOleg O SUSHKOV, Claude SAMMUTThe University of New South Wales

We present a method for matching image localfeatures, specifically SIFT features, to a database oflearned object features for the purpose of objectrecognition and localisation. Our approach differsfrom existing methods by taking into account thegeometric consistency of matched featuresconcurrently with their description vector similarity.As a result we do not need to over-constrain thedescription vector matching criteria (descriptionvectors of matching features do not need to benearest neighbours). The outcome of our approach isa greater number of feature matches between ascene image and a database image, as well animprovement in matching speed under certaincircumstances.

TI.24 P0437Human Face Detection using Color Spaces andRegion Property MeasuresVijayanandh R, *BALAKRISHNAN GJ.J. College of Engineering and Technology*Indra Ganesan College of Engineering

Face detection is the first step in automated facerecognition and the important cue to detect thehuman face is skin color. This proposed paperconsists of two steps, the first step is to detect themore skinned region by the fusion of RGB, YCbCr skincolor region and CIEL*a*b skin labeled image usingHillClimbing segmentation with K-Means clustering.This step helps to simplify and improves the detectionof human faces, since nonskin regions are eliminated.The second step is to locate the faces in a fusedimage by Region property measures. The proposedmethod has been tested on various real images andits performance is found to be quite satisfactory.

TI.25 P0609An Improved Algorithm for Segmenting andRecognizing Connected Handwritten

CharactersXiaoyu ZHAO, Zheru CHI, Dagan FENGThe Hong Kong Polytechnic University

In this paper, an improved algorithm is proposedfor the segmentation and recognition of handwrittencharacter strings. In the method, a gradient descentmechanism is used to weigh the distance measure inapplying KNN for segmenting/recognizing connectedcharacters (numerals and Chinese characters) in theleft-to-right scanning direction. In recognizingconnected characters, a high quality segmentationtechnique is essential. Conventional approachesattempt to separate the string into individualcharacters without recognition and apply arecognition algorithm onto each isolated character,resulting improper segmentation and poor recognitionresults in many situations. Our proposed algorithmsimulates the human beings's process in recognizingconnected character strings where segmentation andrecognition is mingled with each other. Experimentalresults on 1959 character strings from the USPSdatabase of postal envelopes show that the algorithmworks robustly and efficiently.

TI.26 P0790Automatic Traffic Monitoring using NeuralNetworks from Satellite ImagesMehrad ESLAMI, *Karim FAEZAzad University of Qazvin*Amirkabir University of Technology, Tehran

Considering the widespread problems of roadtransport, approach of the paper is a system toautomatically control the roads by using images fromsatellite in night and day. Although no coherentsystem with appropriate performance has been yetintroduced to achieve this goal, some methods hasbeen proposed to estimate the road or recognizeobjects on the road, which have been more based onthresholding and color-based object recognition; andtherefore, their efficiencies have direct relationshipand a lot of dependence with the type of inputimage. In this paper, a complete and coherentsystem has been introduced to detect traffic by usingsatellite images, in which a special attention is paid toextraction of the road and vehicles on the road byusing image processing and machine learning(including feature extraction, morphology methods,and algorithms of labeling); and rate of road traffic isestimated by using the obtained results and by usingneuro-fuzzy network. Previous works has beenintroduced in the paper; and finally, the obtainedresults have been compared with the pastappropriate methods. Higher accuracy and lessdependence on the input image is among the resultsthat have been explained in detail in last section ofthe paper in which and results for various imagesfrom satellite show an accuracy of about 85%.

TI.27 P0832Sequential Monte-Carlo Estimation Based RoadRegion Segmentation Considering PerspectiveProjection

92

Zdenek PROCHAZKAOita National College of Technology

Road region is an important information forguidance of such objects like robots or autonomousvehicles. The goal of this research is to develop arobust monocular algorithm for extraction of roadregion from image sequences, captured by cameramounted on the moving objects. The key ideaintroduced in the first stage of this research is toexpress the road region in terms of probabilitydensity function (pdf) of color and spatial values, andrecursively estimate the pdf by sequential Monte-Carlo estimation. This paper describes an extensionto the proposed method in the sense that newlyconsider perspective projection of road plane bycamera. The proposed approach is examined on realimage sequences, and compared with the first one.

TI.28 P0845An Efficient Edge and Corner DetectorSong YANG, Siew Kei LAM, Thambipillai SRIKANTHANNanyang Technological University

This paper describes an efficient method for imageedge and corner detection. Edges are detectedbefore extracting corner points so that thebackground or noise of the image can be effectivelyremoved. We propose an edge detector that utilizesedge features to localize the edge points. The cornerpoints of an image are then identified by evaluatingthe degree of turning in the edge direction. Theproposed edge and corner detection utilizes lowcomplexity computations. Based on the experimentsconsidered, it can be observed that the proposededge and corner detector achieves more accurateresults than several other well known algorithms. Inaddition, the proposed method performs faster thanmost of the other well known algorithms.

TI.29 P1028An Initial Edge Point Selection and SegmentalContour Following for Object ContourExtractionRoy Chaoming HSU, Ping-Wen KAO, Wei-Jie LAI,Cheng-Ting LIUNational Chiayi University

Image segmentation is a technique used tosegment region of interest (ROI) in the image forfurther image processing applications. Contourextraction is one of the most important imagesegmentation methods. In this paper, a contourextraction method is proposed, which consist of twostages. The first stage is automatic initial edge pointselection (IEPS), and the second stage is segmentalcontour following (SCT) based on the edge pointsobtained at the first stage. Experiment result showsthat, in comparing with other existing methods, moreaccurate contour can be obtained with lowcomputational complexity using the proposed objectcontour extraction method, which will benefit toapplications of industrial inspection or machine visionwith low-level processor.

TI.30 P1130Advance in Triangular Mesh SimplificationStudyMingyi HE, Long LINorthwestern Polytechnical University

With the development of modern 3D acquisitionfacilities and tools, large mesh models with high-precision for the representation of complex geometricobjects becomes feasible. However, manyconsiderable difficulties have emerged in these hugemesh models, such as the storage capacity andrendering speed of computers. Triangular mesh,which is one of the most popular polygonal meshes,plays an important role in mesh simplification field.Many different methods and algorithms have beenput forward to the research of triangular meshsimplification in the past few years. In this paper,triangular mesh simplification algorithms and theircorresponding improved algorithms are overviewed.In addition, some recent achievements of triangularmesh simplification in the author's laboratory (IAP)are presented. Finally, future development andprospects in this area are also discussed andoutlined, respectively.

TI.31 P1001Facial Part Displacement Effect on Template-based Gender and Ethnicity ClassificationFahimeh SAEI MANESH, Mohammad GHAHRAMANI,Yap-Peng TANNanyang Technological University

Visual information such as gender, age andethnicity play critical roles in human identification.Most of gender and ethnicity recognition researchworks use the full face considering equal discriminantcapability to different face parts. In this paper, weimprove the gender and ethnicity recognition, byemploying the optimum decision making rule on theconfidence level of automatically separated faceregions using the modified Golden ratio mask. Facesare preprocessed with multiple base pointphotometric normalization to prevent thedisplacement of facial parts in the noted mask, due todifferent facial parts’ distances of the people. SVM isemployed as the classifier on the extracted Gaborfeatures of each patch to get its confidence level. Thefinal classification results are obtained based on theoutput of each patch decision using the optimumdecision making rule. Finally, using the most accuratenormalization approach for each patch, we couldachieve 94% and 98% for gender and ethnicityrespectively on a dataset composed of FERET andPEAL frontal face images.

TI.32 P0407Fluid Dynamics Analysis of Contrast Media inLeft VentriculographyLay-Lan LEE, *Gwo-Chung TSAI, **Chien-Wei LIUTzu Chi University*National I-Lan University

93

**St. Mary's CollegeA contrast media was injected into the left

ventricle through a pig-tail catheter and imaged withX-ray angiography to observe pathological changes ofthe left ventricle. The sidehole design of the cardiacpigtail catheter affects scope visualization, andlimited research related to this subject has beenperformed. In this study, a computational fluiddynamics method was applied to observe thedistribution of Renografin-76 contrast media usingfive and eight sideholes cardiac catheters. The resultsshowed that the eight sideholes catheter had a morerapid rate of contrast media delivery compared to thefive sideholes catheter. In conclusion, a larger area ofthe left ventricle was observed through the eightsideholes catheter compared to the five sideholescatheter.

TI.33 P0580A Patch-based Spatiotemporal PhaseUnwrapping Method for Phase Contrast MRIusing Graph CutsWenyu XIE, Ying SUN, Sim Heng ONGNational University of Singapore

Phase unwrapping is an important and challengingproblem in phase contrast magnetic resonanceimaging (PC-MRI). In this paper, we propose a newalgorithm for phase unwrapping based on graph cuts.Our algorithm takes a patch-based approach whichhas the advantages of simplicity and robustness tophase noise. To make use of temporal informationfrom the neighboring frames as well as spatialinformation from the current frame, the energyfunction is designed to combine both spatial andtemporal constraints. The proposed method has beentested with real PC-MRI data. Experimental resultsdemonstrate that our algorithm is capable ofunwrapping images with severe phase wrapping, andthat it outperforms other existing popular phaseunwrapping algorithms both qualitatively andquantitatively.

Session FA1Date : Friday, 10 December 2010Time : 1000 - 1200Venue : Ball Room

Nonlinear SystemsChairs : Bill GOODWINE

University of Notre DameMaryam DEHGHANIShiraz University

FA1.1 P0147 1000 - 1020Robust Stabilization using Time-scaling andLyapunov Redesign: The Ball-beam SystemR Maruthi T, Arun D MAHINDRAKARIndian Institute of Technology, Madras

This paper proposes a nonlinear stabilizingcontroller for a ball on an end-actuated beam system,which is robust to an uncertainty in the mass of theball. To this end, the dynamics is time-scaled into twosubsystems termed as the ‘Outer-Loop' and the`Inner-Loop' dynamics. An Outer-Loop controllergenerates a reference trajectory for the beam's pitchangle, which if faithfully followed, would result in thestabilization of all states of the system. A robustInner-Loop controller is synthesized using theLyapunov redesign technique, which forces the actualpitch angle to stabilize to the trajectory of the Outer-Loop controller. It is shown that the effects of theuncertainty in the mass of the ball are eliminated bythe Inner-Loop controller. The uncertainty tolerancelimit of this robust controller is also characterizedthrough a necessary condition. Experiments validatethe effectiveness of this strategy in stabilizing thesystem.

FA1.2 P1029 1020 - 1040Lyapunov Based H∞ Controller Design inMultimachine Power SystemsMaryam DEHGHANI, *Seyyed KamaleddinNIKRAVESHShiraz University*Amirkabir University of Technology, Tehran

In this paper, nonlinear H∞ controller problem inlarge-scale systems is investigated. The paper needsto find an energy function which satisfies theHamilton Jacobi inequality. The energy function ofthe whole large scale system is considered to be thesum of the energy functions of the isolatedsubsystems. When the inequality is solved, adecentralized state feedback controller for eachsubsystem is obtained. The controller can beimplemented locally in each subsystem. The methodis tested on a sample multi-machine power systemmodel. Simulation results show the effectiveness,robustness and good performance of the proposedcontroller.

94

FA1.3 P1096 1040 - 1100Multi Objective Evolutionary Programming toSolve Environmental Economic DispatchProblemBo-Yang QU, Suganthan PONNUTHURAINAGARATNAM, *V. R. PANDI, *B. K. PANIGRAHINanyang Technological University*Indian Institute of Technology, Delhi

In this paper, the nonlinear constrained multi-objective environmental economic dispatch (EED)problem is solved using fast multi-objectiveevolutionary programming (FMOEP). Due to theglobal warming by fossil fuel, environmental concernbecomes more and more important in recent years.The purpose of multi-objective optimization algorithmis minimizing all the different objectivessimultaneously and finds the best tradeoff solutionfor this environmental/economic dispatch problem. Inorder to evaluate the performance of FMOEP on EEDproblems, the standard IEEE 30-bus six-generatortest system is studied. The performance is comparedagainst NSGAII and a number of results reported inliterature. The results show that the FMOEP iseffective in solving EED problems.

FA1.4 P0231 1100 - 1120Adaptive Control of Singular NonlinearSystems with Convex/ConcaveParametrizationQingxiang FANG, Feilong CAOChina Jiliang University

This note is concerned with the model referenceadaptive tracking problem of singular nonlinearsystems with nonlinear parametrization where thenonlinearity in the parameters is convex or concave.Applying the standard coordinate transformation tothe singular system to yield a reduced-order normalsystem, we convert the model reference adaptivetracking problem of singular system into the modelreference adaptive tracking problem of normalsystem and obtain solvability conditions. Theproposed controller ensures that the overall adaptivesystem has globally bounded solutions and achievestracking to within a desired precision.

FA1.5 P0278 1120 - 1140Multi-layer Moving-window Hierarchical NeuralNetwork for Modeling of High-densityPolyethylene Cascade Reaction ProcessYuan XU, Qunxiong ZHUBeijing University of Chemical Technology

With the growing scale of industry production,process modeling has been paid more and moreattention, which could effectively explore thedynamics of the process and provide guidelines toproduction operation. High-density polyethylene(HDPE) cascade reaction process is such a complexand nonlinear industry process. To enhance theperformance of process modeling, a multi-layermoving-window hierarchical neural network(MMHNN) is proposed, which is developed with the

incorporation of multi-layer moving-window conceptand hierarchical neural network (HNN). Multi-layermoving-window is used to ensure the continuity andtime-variation, HNN is used for input compressionand model prediction, which can effectively capturethe changing process dynamics, reduce the datadimension and reveal the nonlinear relationshipbetween process variables and final output. Forcomparison, single-layer moving-window HNN(SMHNN) and HNN are also established for theprocess modeling. Through the actual application inHDPE cascade reaction process of a chemical plant,the prediction results show that MMHNN is obviouslybetter than SMHNN and HNN with higher accuracy,thus exploits a new and efficient way to simulate andguide the industry process.

FA1.6 P0757 1140 - 1200Vehicle Formation Analysis based on ZeroDynamics for Road Traffic SystemDi-Hua SUN, Yong-Fu LIChongqing University

Car-following models play an important role in thefield of road traffic system, and one of its importantissues is the vehicle formation. To conduct theanalysis of vehicle formation, this paper introducesthe concept of zero dynamics and gives a detailedanalysis of the full velocity difference (FVD) model.Meanwhile, the stability analysis of FVD model basedon the Lyapunov function is also discussed, and theresult validates that the vehicle formation can achieveasymptotic stability.

95

Session FA2Date : Friday, 10 December 2010Time : 1000 - 1200Venue : Canary

Networked Control SystemsChairs : Young Soo SUH

University of UlsanGang ZHANGHebei Normal University

FA2.1 P0371 1000 - 1020Synchronization of Networked HarmonicOscillators under Nonlinear ProtocolsShan CHENG, *Gang ZHANG, Lan XIANG, Jin ZHOUShanghai University*Hebei Normal University

This paper deals with the synchronization ofnetworked harmonic oscillators under nonlinearprotocols. Some convergence analysis for suchnetworked harmonic oscillators, both without andwith a leader, are provided by means of algebraicgraph theory, matrix theory and Lyapunov stabilitytheory on dynamical systems. We show that thenetworked harmonic oscillators can achievesynchronization under nonlinear control input.Moreover, we also demonstrate that all harmonicoscillators can attain the position and velocity of theleader under nonlinear network connectivityconditions. Finally, numerical simulations are given toverify the correctness of obtained results.

FA2.2 P0934 1020 - 1040Inertial Sensor Data Compression usingModified ADPCMYoung Soo SUH, Young Shick RO, Hee Jun KANGUniversity of Ulsan

This paper is concerned with compression ofmultiple inertial sensor data. The compressed dataconsists of ADPCM compressed data and an adaptiverefinement data. Each sensor data is first compressedusing ADPCM. To reduce the quantization errors,refinement quantizers are then used, where the totalnumber of bits for the refinement quantizers is fixed.Which inertial sensor data should be further refined isadaptively determined based on quantization errorbounds. In the proposed method, total number ofbits for the compressed data is fixed; however thenumber of bits for individual inertial sensor isadaptively determined. Through experiments, it isshown that the proposed method provides bettercompression quality than the standard ADPCM withthe same total number of bits.

FA2.3 P0987 1040 - 1100Convergence of Kalman Filter with QuantizedInnovationsJian XU, Jianxun LI, *Jiayun WU

Shanghai Jiaotong University*National University of Singapore

This work provides a convergence analysis for theestimate error covariance of Kalman filtering basedon quantized measurement innovations (QIKF). Bytaking the quantization errors as randomperturbations in observation system, an equivalentstate-observation system is given. Accordingly, thequantitative Kalman filter for the original system isequivalent to a Kalman-like filtering for the equivalentstate-observation system. In this performanceanalysis framework, the true covariance matrix ofestimating error is strictly analyzed without Gaussianassumption on predicted distribution. A necessaryand sufficient condition for the stability of the QIKF isobtained. Then, the relationship between thestandard Kalman filtering and the QIKF for theoriginal system is discussed. Finally, the validity ofthese results are demonstrated by numericalsimulations.

FA2.4 P1023 1100 - 1120Security Enhancement with Optimal QoS usingECDH for Converged 3G-WLAN SystemShankar RAMALINGAM, *Timothy RAJKUMAR K,*Dananjayan PERUMALSri Manakula Vinayagar Engineering College,Pondicherry*Pondicherry Engineering College

The convergence of third generation (3G) widearea wireless network and wireless local area network(WLAN) offers characteristics that complement eachother perfectly. To provide secure 3G-WLANconvergence, extensible authentication protocol-authentication and key agreement (EAPAKA)protocol, that has got many risk, is used. This paperelaborate risks in using EAP-AKA for 3G-WLANconvergence, proposes an authentication and keyagreement protocol that combines elliptic curveDiffie-Hellman (ECDH) with symmetric keycryptosystem. The proposed protocol when analysedusing simulation tool shows that it not onlyovercomes the risk, it ensures optimal quality ofservice (QoS) because of its less delay forauthentication compared to EAP-AKA

FA2.5 P0248 1120 - 1140Simple Analog-Digital Circuit for Detection ofApproaching Object based on Visual Systemsof Pigeon and LocustKimihiro NISHIO, Naoya IHARA, Takumi YAMASAKITsuyama National College of Technology

We proposed in this study the circuit and networkfor detecting the approaching object based on thevisual systems of the pigeon and the locust. Theproposed circuit and network became simplestructure. The circuit and network were evaluated bythe simulation program with integrated circuitemphasis (SPICE). The circuit based on the visualsystem of the pigeon output the signal which isproportional to the time to collision. The circuit and

96

network based on the locust vision system generatedthe signal for avoiding the collision of the object. Itwas clarified from the simulation results that theproposed simple circuit and network can detect thetime to collision, the size and the velocity of theapproaching object.

FA2.6 P0467 1140 - 1200Incremental Learning Bayesian NetworkStructures EfficientlyShi DA, Shaohua TANPeking University

In this paper, a new hybrid incremental learningalgorithm for Bayesian network structures isproposed. It develops a polynomial-time constraint-based technique to build up a candidate parents setfor each domain variable, and a hill climbing searchprocedure is then employed to refine the currentnetwork structure under the guidance of thosecandidate parents sets. Our algorithm always offersconsiderable computational complexity savings whileobtaining better model accuracy compared to existingincremental algorithms when dealing with complexreal-world problems. The more complex the real-world problems are, the more significant theadvantage our algorithm keeps is.

Session FA3 - Invited SessionDate : Friday, 10 December 2010Time : 1000 - 1200Venue : Kingfisher

Robotics in Unstructured EnvironmentsChairs : XiaoQi CHEN

University of CanterburyDe XUChinese Academy of Sciences

FA3.1 P0308 1000 - 1020Wavelets/Multiwavelets Bases andCorrespondence Estimation Problem: AnAnalytic StudyAsim BHATTI, Saeid NAHAVANDI, MohammedHOSSNYDeakin University

Correspondence estimation in one of the mostactive research areas in the field of computer visionand number of techniques has been proposed,possessing both advantages and shortcomings.Among the techniques reported, multiresolutionanalysis based stereo correspondence estimation hasgained lot of research focus in recent years.Although, the most widely employed medium formultiresolution analysis is wavelets and multiwaveletsbases, however, relatively little work has beenreported in this context. In this work we have tried toaddress some of the issues regarding the work donein this domain and the inherited shortcomings. In thelight of these shortcomings, we propose a newtechnique to overcome some of the flaws that couldhave significantly impact on the algorithm perfor-mance and has not been addressed in the earlierpropositions. Proposed algorithm uses multiresolutionanalysis enforced with wavelets/multiwaveltstransform modulus maxima to establishcorrespondences between the stereo pair of images.Variety of wavelets and multiwavelets bases,possessing distinct properties such as orthogonality,approximation order, short support and shape areemployed to analyse their effect on the performanceof correspondence estimation. The idea is to provideknowledge base to understand and establishrelationships between wavelets and multiwaveletsproperties and their effect on the quality of stereocorrespondence estimation.

FA3.2 P0313 1020 - 1040Control System Design for a 5-DOF TableTennis RobotPing YANG, De XU, Huawei WANG, Zhengtao ZHANGChinese Academy of Sciences

A visual control system is designed for a tabletennis robot with five degrees of Freedom (DOFs). Itconsists of four parts such as ball sensing, trajectorypredicting, motion planning, and motion control. Ahigh-velocity stereo vision system with parallelarchitecture is developed to sense the motions of

97

table tennis ball. The striking parameters includingposition, velocity, and time are predicted according tothe predicted trajectory of the ball based on severalmeasured positions. The motion computer receivesthe predicted striking parameters and performsmotion planning for the robot. A motion control cardembedded in the motion computer receives theplanning results and controls the motions of the robotvia the servo drivers for X and Y axes. Amicroprocessor is designed to produce pulses tocontrol the motions of the rest three axes via thedrivers. Experiments are well conducted to verify theeffectiveness of the developed robot and controlsystem.

FA3.3 P0360 1040 - 1100Rocket Roll Dynamics and Disturbance -Minimal Modelling and System IdentificationChristopher E HANN, Malcolm SNOWDON, AvinashRAO, Robert TANG, Agnetha KOREVAAR, GregSKINNER, Alex KEALL, et alUniversity of Canterbury

The roll dynamics of a 5kg, 1.3 m high soundingrocket are analyzed in a vertical wind tunnel.Significant turbulence in the tunnel makes the systemidentification of the effective inertia, damping andasymmetry with respect to roll challenging. A novelmethod is developed which decouples thedisturbance from the rocket frame’s intrinsic rolldynamics and allows accurate prediction of roll rateand angle. The parameter identification method isintegral-based, and treats wind disturbances asequivalent to a movement in the actuator fins. Themethod is robust, requires minimal computation, andgave a realistic disturbance distribution reflecting therandomness of the turbulent wind flow. The meanabsolute roll rate of the rocket frame observed inexperiments was 16.4 degree/s and the modelpredicted the roll rate with a median error of 0.51degrees/s with a 90th percentile of 1.25 degrees/s.The roll angle (measured by an encoder), wastracked by the model with a median absolute error of0.25 degrees and a 90th percentile of 0.50 degrees.These results prove the concept of this minimalmodeling approach which will be extended to pitchand yaw dynamics in the future.

FA3.4 P0492 1100 - 1120Spline-Interpolation Based PVT Algorithm andApplication in a Bionic Cockroach RobotHaosong YUE, Weihai CHEN, *Wenjie CHEN,Xingming WUBeijing University of Aeronautics and Astronautics*Singapore Institute of Manufacturing Technology

To solve the problem that the final velocity curvewill be unsmoothed if the PVT(Position Velocity Time)nodes velocities are determined improperly whenusing PVT control algorithm, based on analyzing thegeneral principle of spline interpolation and PVTmotion control, this paper presents a triple splineinterpolation based PVT algorithm, which ensures thesmoothness of the velocity curve. The velocity of

each node is firstly calculated via spline interpolation,and then the final trace control curve is achieved byusing PVT method to these points. The proposedalgorithm is applied to the kinematics control of acockroach robot. The structure and forward/inversekinematics model of the cockroach robot isexpounded. Finally, the reliability of the proposedalgorithm is verified by the contrast simulation andexperiment of a single leg of the cockroach robotwith one of the existing algorithms. The results showthat the proposed approach can be readily used formotion controls.

FA3.5 P0528 1120 - 1140Towards Autonomous Image FusionMohammed HOSSNY, Saeid NAHAVANDI, DougCREIGHTON, Asim BHATTIDeakin University

Mobile robots are providing great assistanceoperating in hazardous environments such as nuclearcores, battlefields, natural disasters, and even at thenano-level of human cells. These robots are usuallyequipped with a wide variety of sensors in order tocollect data and guide their navigation. Whether asingle robot operating all sensors or a swarm ofcooperating robots operating their special sensors,the captured data can be too large to be transferredacross limited resources (e.g. bandwidth, battery,processing, and response time) in hazardousenvironments. Therefore, local computations have tobe carried out on board the swarming robots toassess the worthiness of captured data and thecapacity of fused information in a certain spatialdimension as well as selection of proper combinationof fusion algorithms and metrics. This paperintroduces to the concepts of Type-I and Type-IIfusion errors, fusion capacity, and fusion worthiness.These concepts together form the ladder leading toautonomous fusion systems.

FA3.6 P0736 1140 - 1200A Flexure-Based 4-DOF Coaxial AlignmentSystem: Design and ApplicationWenjie CHEN, Wei LIN, Guilin YANGSingapore Institute of Manufacturing Technology

This paper presents the design of a novel 4-DOF(x-y-¦Èx-¦Èy) flexure-based system used for coaxialalignment. The movable head (a rigid rod) of thesystem is suspended with two monolithic flexuralmechanisms constructed with leaf flexures. Toachieve the required x-y-¦Èx-¦Èy motion, the leaf-type flexures are specially arranged in such aconfiguration that is compliant in the specified axesbut stiff in other axes. The leaf flexures are initiallydesigned based on the linear models of the flexiblebeam deflection and then improved through FEAsimulations. A system prototype was built with theflexural mechanisms. It demonstrates superiormotion performances in the x and y axes as well asthe angular axes. The prototype has beensuccessfully applied in the coaxial alignment of thecapillary machining.

98

Session FA4Date : Friday, 10 December 2010Time : 1000 - 1200Venue : Nightingale

Intelligent AutomationChairs : Manish M PATIL

MAE College of Engineering, PuneYongduan SONGBeijing Jiaotong University

FA4.1 P0624 1000 - 1020A Novel SVM Based Approach for Noisy DataEleminationNarendra S CHAUDHARI, *Aruna TIWARI, *JayaTHOMASIndian Institute of Technology, Indore*SGS Institute of Technology and Science, Indore

In this paper we propose a novel Support VectorMachine(SVM) based approach for noisy data removalfrom datasets. It is observed that the instabilitypresent in the dataset greatly affects the overallperformance of the any classifier. Hence, we proposea methodology for removal of such instabilities. Inthe proposed approach, we proceed by determiningthe clusters formed using support equilibrium points.Then analyzing, each cluster and remove the noisydata using the accuracy factor. Our approach,provide an important feature for reducing the trainingtime and reducing the misclassification test error. Themethodology if adopted for classifiers before thetraining phase will enhance the efficiency of thesystem. The approach is being tested on benchmarkdataset, and it is observed that the efficiency ofclassifier increased by 15-20%.

FA4.2 P0714 1020 - 1040Novel Simulated Annealing Algorithm in orderto Optimal Adjustment of Digital PIDControllerMohammad DASHTI J, *Kambiz SHOJAEE G, **SMohammad H SEYEDKASHI, ***Mojtaba BehnamTAGHADOSISadjad Institute of Higher Education*University of Tehran**Tarbiat Modares University***Politecnico di Torino

This article presents a method for optimal tuningof PID controllers by using simulated annealingoptimization algorithm and tries to increase speedand decrease errors in PID controller by addingintelligent techniques to the basic algorithm. Theobjective of this paper is to reduce errors andminimize IAE, ITAE, ISE and ITSE assessment criteriato be able to obtain a unique step response of thesystem in its best output tuning by PID controller.This system is not limited to a particular process andis applicable for all plants. In this paper, SMT motionsystem with constant load torque is used as

disturbance for execution of PID controller with thisproposed method.

FA4.3 P0828 1040 - 1100Multi-criteria Optimization of the ParallelMechanism with Actuators Located OutsideWorking SpaceMinh Thanh NGUYEN, *Victor GLAZUNOV, Cong TuanTRAN, Xuan Vinh NGUYENHo Chi Minh City University of Transport*Russian Academy of Sciences

This paper addresses the development ofautomation of designing a parallel mechanism withactuators located outside the working space which inpurpose to achieve an increase their working volumeand load capacity. It is known that a parallelmechanism with singularity configurations can gainone or more degrees of freedom and becomeuncontrollable. That is it might not reproduce a stablemotion under a prescribed trajectory. However, it isproved that there is possible passing through thesingular zones. Therefore, the utility of using theapproach determination of the twists insidesingularity to rely on the theory of screws isconsidered. Besides, the method is used to proposeoptimal design of the parallel mechanisms, based onthe multi-criteria optimization obtaining the Paretoset.

FA4.4 P0869 1100 - 1120Dealing with Nonlinear and UncertainNonlinear Resistances in Train Control viaAdaptive ApproachQi SONG, Yongduan SONGBeijing Jiaotong University

Resistive forces such as mechanical frictions andaerodynamic drags are inevitable in high speed trainduring its operation. This paper investigates theproblem of adaptive compensation of such uncertainresistive forces in train systems to achieve highprecision speed and position tracking. A controlscheme is developed via adaptive backsteppingapproach to address not only the traction and brakingdynamics ignored in most existing methods, but alsothe uncertain resistive forces arisen from varyingoperation conditions. Both theoretical and numericalsimulation confirms that the resultant controlalgorithms are able to achieve high precision trainposition and speed tracking.

FA4.5 P0664 1120 - 1140Fuzzy Feeder Models for Distribution SystemCalculationsShyi-Wen WANGChienkuo Technology University

Using fuzzy set theory, this study presents threenovel distribution fuzzy feeder models to simplifycomplicated distribution system calculations. Thesefuzzy models are formulated using lumpeddistribution transformer(s) and lumped load(s) to

99

represent non-uniformly distributed loads, which varyby time and type of day and relate to voltage, alonga feeder. These equivalent models are developed toaccurately simulate first, the total series voltage dropat the end of given feeder, and then the total copperand core losses. Finally a hybrid model is developedto simulate both voltage drop and feeder lossesaccurately. A physical feeder is used as a samplefeeder to demonstrate the correctness of theseproposed fuzzy feeder models.

FA4.6 P0906 1140 - 1200Novel Integrated Development Environmentfor Implementing PLC on FPGA by ConvertingLadder Diagram to Synthesizable VHDL CodeShaila SUBBARAMAN, *Manish M PATIL, **PrashantS NILKUNDWalchand College of Engineering, Sangli*MAE College of Engineering, Pune**Laboratory for Applied Research in Electronics,Mumbai

Programmable logic controllers (PLCs) are theuniversally accepted automation components inindustrial control. IEC61131-3 standard wasdeveloped to unify textual and graphical ways ofdescribing the control specification for PLCs; ladderdiagram being one such control specifications.However, PLCs available in the market as on todayare vendor specific and replacement of PLC of onevendor in a process equipment by PLC of anothervendor is not possible due to lack of standardizationamong EDA tools from PLC vendors. Secondly, heartof PLCs is a sequential processor which cannotexecute parallel rungs of ladder diagramconcurrently. Perhaps this aspect did not pose anyproblem so far due to large response timesassociated with industrial processes. However withintroduction of high speed MEMS sensors along withincreased control complexity, the need for high speedPLCs has been spelt out. Further, the safety featuresof control system are now expected to be integralpart of ladder control specification. FieldProgrammable Gate Arrays (FPGAs) have betterportability support, ability to support implementationof concurrent logic and security aspects. They havebeen mentioned in the literature as prospectiveprogrammable devices to implement control logic ofPLCs. However, utilization of FPGAs for implementingcontrol specification demands development ofintegrated environment to convert ladder logic as perIEC61131-3 to HDL. This HDL specification furthercan be ported to any of the standard FPGAdevelopment environments. It is seen throughliterature that not enough efforts have been focusedon developing such integrated environment. Thispaper focuses on the details of such environmentbeing developed by authors to convert IEC-61131-3control specification standard to VHDL for directsynthesis.

Session FA5 - Invited SessionDate : Friday, 10 December 2010Time : 1000 - 1200Venue : Pelican

Autonomy of Vehicular Systems IIChairs : Philippe MARTINET

Clermont Universit´eHan WANGNanyang Technological University

FA5.1 P1153 1000 - 1020A GA based SLAM with Range Sensors onlyHan WANG, Yi YAN, Danwei WANGNanyang Technological University

This paper describes a GA approach for solving theSLAM problem. We treat the local scan from lasersensor as an image pattern. Two subsequent scanswere matched to find the robot’s ego motion. Thismotion is used as updates for robot’s global positionestimation in the map. A virtual scan is obtained fromthe map in the robot’s current position. This virtualscan is then matched with the current laser scan toupdate robot’s location in the map. The innovation inthis paper is the matching algorithm to determine therelative location of two patterns: (1) two successivescans and (2) current scan with virtual scans takenfrom the map. The matching algorithm is to find theoptimal location of one pattern to the other, namely(dx,dy,dθ) (two for translation and one for rotation).This is reduced to a three-dimensional searchproblem. The GA we developed can achieve highaccuracy with fast computational time. Experimentsare performed under simulated and real data. Theperformance obtained outperforms ICP.

FA5.2 P1120 1020 - 1040Urban Vehicle Platoon using Monocular Vision:Scale Factor EstimationPierre AVANZINI, Benoit THUILOT, PhilippeMARTINETClermont Universit´e

Environment, sustainable development as well asnew transportation service emergence in urban areasare major concerns. Consequently, studies arecurrently intended to automate electric vehiclesdesigned for applications in free access. An additionalfunctionality that appears very attractive is vehicleplatooning. In order to avoid oscillations within thefleet when completing this task, a global controlstrategy, supported by inter-vehicle communications,is investigated. Vehicle absolute localization is thenneeded and is here derived from monocular vision.These data are however expressed in a virtual visionworld, slightly distorted with respect to the actualmetric one. It has previously been shown that such adistortion can accurately be corrected on-line indifferent ways, considering telemetric or odometricdata. These strategies have here been refined in

100

order to provide optimal corrections. A comparativestudy, supported by simulations and full-scaleexperiments, is reported to exhibit benefits andperformances of proposed approaches.

FA5.3 P1150 1040 - 1100A Lidar Perception Scheme for IntelligentVehicle NavigationJulien MORAS, V´eronique CHERFAOUI, PhillipeBONNIFAITUniversité de Technologie de Compi`egne

In urban environments, moving obstaclesdetection and free space determination are key issuesfor driving assistance systems or autonomousvehicles. This paper presents a lidar-based perceptionsystem able to do simultaneously mapping andmoving obstacles detection. Nowadays, many lidarsprovide multi-layer and multi-echo measurements. Asmart way to handle this multi-modality is to usegrids projected on the road surface in both global andlocal frames. The global one generates the mappingand the local one is used to deal with moving objects.An approach based on both positive and negativeaccumulation has been developed to address theremnant problem of quickly moving obstacles. Thismethod is also well suited for multi-layer and multi-echo sensors since it is able to handle efficiently theuncertainty that exists between the projected echoes.Experimental results obtained with an IBEO Alascaand an Applanix positioning system show theperformance of such a perception strategy.

FA5.4 P0523 1100 - 1120Design of a Wheel-Propeller-Leg IntegratedAmphibious RobotJiancheng YU, Yuangui TANG, Xueqiang ZHANG,Chongjie LIUChinese Academy of Sciences

The operation capabilities of robot in theamphibious environments (such as shallow waterfields, surf zones, and beaches) are critical formilitary and civilian. In this paper, we introduce anovel amphibious robot with wheel-propeller-legintegrated driving devices, developed by ShenyangInstitute of Automation, which can realize bothcrawling locomotion on the ground and swimminglocomotion in the water without changing its drivingdevices. This paper describes the design of theoverall robot structure, the design of the noveldriving devices, and the design of the embeddedcontrol system, respectively. All the driving devices ofthe robot are driven by independent motor, thus theamphibious robot can conveniently switch itslocomotion modes according to the operationalenvironments. The embedded control system is adistributed control system based on CAN bus, whichmakes it is easy to expend sensors and devices forthe robot in the future. Finally, the hydrodynamicperformances of the wheel-propeller device areanalyzed by using CFX hydrodynamic calculationsoftware, and some primary experiments have been

done for verifying the fundamental locomotionfunctions of the robot.

FA5.5 P0352 1120 - 1140A MDL-Based Control Method for Tele-roboticSystems over InternetJianning HUA, *Hongyi LI, *Yuechao WANG, *Ning XINortheastern University*Chinese Academy of Sciences

A MDL-based control method for tele-roboticsystems over internet is presented in this paper.Internet-based tele-robotic systems are characterizedby the fact that the operator and the remote roboticsystem are connected by Internet. Since limitedbandwidth, random time delay and othertransmission problems deteriorate performance oftele-robotic systems severely, control strategieswhich can be used to settle transmission problemsare expected. The obvious characteristic of the newmethod introduced in this paper is that controlcommands have a linguistic flavor, data transferredbetween operator and remote robot can be reducedaccordingly. The framework presented in this paperhas the potential of reducing operator’s workingpressure and enhancing the performance of thesystem. Furthermore, the existing motion descriptionlanguage modal is expanded to deal with the packetloss and disorder problem. The main contribution ofthis paper is the development of a novel frameworkfor tele-robotics and its implementation in a real tele-manipulator system.

FA5.6 P0516 1140 - 1200Development of a Tele-guidance System withFuzzy-based Secondary ControllerManh Duong PHUNG, Thanh Van Thi NGUYEN, CongHoang QUACH, Quang Vinh TRANVietnam National University, Hanoi

Dealing with the uncertainties of Internetcharacteris-tics is an important issue that needs beingtaken into account in developing Internet-based real-time systems. In this paper, we present our approachin applying fuzzy logic to develop back-upmechanisms for an Internet-based mobile robot todeal with unwanted network problems such as longdelays or network interruptions. A tele-guidanceapplication involving the remote control of a mobilerobot via the Internet is set up as the context toverify the effectiveness and applicability of theproposed approach.

101

Session FA6Date : Friday, 10 December 2010Time : 1000 - 1200Venue : Oriole

BiometricsChairs : Eric SUNG

Nanyang Technological UniversityXiaozheng ZHANGNational ICT Australia

FA6.1 P0672 1000 - 1020Comparing Human Faces using Edge WeightedDissimilarity MeasureAditya NIGAM, Phalguni GUPTAIndian Institute of Technology, Kanpur

This paper proposes a dissimilarity measure thatcan be used as a distance between two images. Ithas shown better discriminative power to recognizefaces than similar existing variants for discriminatingfacial images. It gives more weight to the pixelswhich are often a part of the edge and counts pixelsthat are unmatched between query and databaseimages. This measure has been tested on a publiclyavailable database as ORL,YALE, CALTEC, BERN andalso on a database developed at IIT Kanpur.Experimental results show that the proposed measureachieves a high recognition rate of 99.75 93.75 99.0398.93 99.73 for the first likely matched faces ondatabases ORL, YALE, BERN, CALTECH, IITKrespectively. The proposed measure can provide notonly effective result against pose and expressionvariations but also against slight illuminationvariation.

FA6.2 P0937 1020 - 1040Pose-Invariant 2.5D Face Recognition usingGeodesic Texture WarpingFarshid HAJATI, Abolghasem A. RAIE, *YongshengGAOAmirkabir University of Technology, Tehran*Griffith University

In recent years, 3D face recognition has become apopular solution to deal with the problem of pose-invariant face recognition. The majority of 3D facedata are, however, actually 2.5D which are sensitiveto pose variations. This paper presents a novelGeodesic Texture Warping (GTW) solution for 2.5Dpose-invariant face recognition. In this method, weuse the geodesic distance computed on a 2.5D facescan to warp the texture of a rotated face to that of afrontal one to perform matching. A feasibility andeffectiveness investigation for the proposed methodis conducted using a wide range of experimentsincluding samples with different face rotations. Theencouraging experimental results demonstrate thatthe proposed method achieves much higher accuracythan the state-of-the-art method with a lowcomputational cost.

FA6.3 P0171 1040 - 1100LBP-Based Biometric Hashing Scheme forHuman AuthenticationZhengyao BAI, *Dimitrios HATZINAKOSYunnan University*University of Toronto

Biometric hash finds extensive applications inmultimedia security systems. Biometric hashingschemes combine biometric features with randomnumbers for robust and secure human authenticationor recognition. A novel biometric hashing schemewhich is secure and robust to lighting changes isproposed in this paper. First, the local binary pattern(LBP) based histogram sequence or vector isemployed to represent a normalized face image.Secondly, the pseudorandom sequence is generatedby using the user specific secret seed (hash key) andit is orthonormalized by using Gram-Schmidtalgorithm. Third, the inner product between thehistogram vector and pseudorandom sequence iscomputed. The biometric hash is obtained bythresholding the inner product with the thresholdcalculated by using Otsu method. Our scheme istested on the face image data. Hamming distance isused to measure the performance of the scheme.Experimental results show that our scheme is secureand robust to lighting changes.

FA6.4 P0605 1100 - 1120Finding Distinctive Facial Areas for FaceRecognitionCe ZHAN, Wanqing LI, Philip OGUNBONAUniversity of Wollongong

One of the key issues for local appearance basedface recognition methods is that how to find the mostdiscriminative facial areas. Most of the existingmethods take the assumption that anatomical facialcomponents, such as the eyes, nose, and mouth, arethe most useful areas for recognition. Other moreelaborate methods locate the most salient partswithin the face according to a pre-specified criterion.In this paper, a novel method is proposed to identifythe discriminative facial areas for face recognition.Unlike the existing methods that only analyze thegiven face, the proposed method identifies thedistinctive areas of each individual’s face by itscomparison to the general population. In particular,nonnegative matrix factorization (NMF) is extended tolearn a localized non-overlapping subspacerepresentation of the facial patterns from a genericface image database. In the learned subspace, thedegree of distinctiveness for any facial area ismeasured depends on the probability of this area isbelong to a general face. For evaluation, theproposed method is tested on exaggerated faceimages and applied in exiting face recognitionsystems. Experimental results demonstrate theefficiency of the proposed method.

FA6.5 P0812 1120 - 1140

102

Recognizing Face Profiles in the Presence ofHairs/glasses InterferencesWeiping CHEN, Yongsheng GAOGriffith University

Facial profile provides a complementary structureof the face that is not present in frontal faces, whichhas been used in personal identification, faceperception research and 3D face construction. In thispaper, we present a novel local attributed stringmatching (LAStrM) approach to recognize faceprofiles in the presence of interferences. Theconventional profile recognition algorithms heavilydepend on the accuracy of the facial area cropping.However, in realistic scenarios the facial area may bedifficult to localize due to interferences (e.g., glasses,hairstyles). The proposed approach is able toefficiently find the most discriminative local partsbetween face profiles addressing the recognitionproblem with interferences. Experimental results haveshown that the proposed matching scheme is robustto interferences compared against several primaryapproaches using two profile image databases (Bernand FERET). It has potential capability for partiallyoccluded shape classification.

FA6.6 P0586 1140 - 1200Dense SIFT and Gabor Descriptors-Based FaceRepresentation with Applications to GenderRecognitionJian-Gang WANG, Jun LI, Chong Yee LEE, Wei YunYAUInstitute for Infocomm Research

In this paper, a novel face representation in termsof dense local image descriptors is proposed. ScaleInvariant Feature Transform (SIFT) and Gabor, twoof the most popular local image descriptors, at densegrid pixels of a face image are used to represent theface. The efficiency of the representation has beeninvestigated in gender recognition. There are fourproblems when applying the SIFT to facial genderrecognition. (1) There may be only a few keypointsthat can be found in a face image due to the missingtexture and ill-illumined faces. (2) The SIFTdescriptors at the keypoints (we called it sparse SIFT)are distinctive whereas alternative descriptors at non-keypoints (e.g. grid) could cause negative impact onthe accuracy. (3) Most of the existing methodsemploy SIFT descriptors matching in which relativelarger image size is required in order that enoughkeypoints can be found to support the matching. (4)The matching is assumed that the faces are wellregistered. We provide solutions to the abovedifficulties in this paper and the problem ofrecognizing gender using the combination of SIFTdescriptors and Gabor of face images is studied. TheGabor representations of the face images are fusedwith the dense SIFT at the feature-level to improvethe accuracy. AdaBoost is adopted to select featuresand form a strong classifier. The experimental resultson a large set of faces have shown that the proposedmethod can achieve high accuracies even for facesthat are not aligned.

Session FA7Date : Friday, 10 December 2010Time : 1000 - 1140Venue : Bluebird

Image/Video Analysis IIChairs : Eam Khwang TEOH

Nanyang Technological UniversityYongqiang YENanjing University of Aeronautics andAstronautics

FA7.1 P0490 1000 - 1020A Neural Network Model with AdaptiveStructure for Image AnnotationZenghai CHEN, Hong FU, Zheru CHI, Dagan FENGThe Hong Kong Polytechnic University

A neural network model with adaptive structure forimage annotation is proposed in this paper. Theadaptive structure enables the proposed model toutilize both global and regional visual features, aswell as correlative information of annotated keywordsfor annotation. In order to achieve an approximateglobal optimum rather than a local optimum, bothgenetic algorithm and traditional back-propagationalgorithm, are combined for model training. Theneural network model is experimented on a syntheticimage dataset with controllable parameters, whichhas not been used in previous image annotationexperiments. Experimental results demonstrate theeffectiveness of the proposed model.

FA7.2 P0769 1020 - 1040Novel Gray-scale Watermarking Algorithmbased on QFTChi ZHAO, Weijiang WANG, *Xiwei HUANGBeijing Institute of Technology*Nanyang Technological University

This paper proposes a novel algorithm ofembedding gray-scale watermark image into an RGBhost image based on the Quaternion FourierTransform (QFT). The R, G, B color components arerepresented respectively by the three imaginary partsof a pure imaginary quaternion and the gray-scalewatermark image is inserted into the real part of thecarrier image which is transformed by the QFFT. Byusing the symmetrical characteristic of the real part,we achieve the insertion of the gray-scale watermarkimage without lose. Experiment results show that thewatermarked image with this method has the biggestimperceptibility with the best PSNR when comparedwith other gray-scale watermarking schemes.

FA7.3 P0879 1040 - 1100Vision-Based Lane Departure Detection Systemin Urban Traffic ScenesYu-Chi LENG, Chieh-Li CHENNational Cheng-Kung University

103

A lane-departure detection system without intrinsicand extrinsic camera parameters calibration isproposed in this paper. The proposed system, whichprovides driving safety by lane detection and lanedeparture warning, is focused on urban road withcomplicated lane marks instead of simple highwayscenes. Due to the complexity of urban traffic scenes,false lane detections are highly caused by warninglines and signs whose shape and color are similar tothe lane boundary. In this study, two lane features,lane width and lane boundary continuity, areproposed to obtain reliable and quality lane detectedresults on urban traffic scenes. Lane departurewarning decisions are then applied to provide drivingsafety. Experimental results of urban road scenerieshave demonstrated feasibility and robustness of theproposed method.

FA7.4 P0916 1100 - 1120A Method for the Segmentation of ConnectedHandwritten Persian DigitsFattaneh TAHERI ASHTIYANI, *Karim FAEZIslamic Azad University, Bonab*Amirkabir University of Technology, Tehran

This article presents in two modules a new methodfor segmenting connected handwritten Persian digitsusing the characteristics of the foreground andutilizing the background skeleton. The first moduleexcavates all the valleys and hills , if there are any,from the upper pixels and lower pixels of the thinnedimage respectively. Then feature point excavate. Forbetter segmentation the digits , a separability degreeis calculated, regarding the height of the hills andvalleys close to each feature point considering thecharacteristic of connected Persian digits. Then thesignificance degree is calculated to determine of itsinfluence rate in segmentation. Then, using that, oneor few points with high significance degree, which aremore influence in the segmentation, are selected asthe cutting points. Having excavated the backgroundskeleton, the second module begins to identify thepriority points in the skeleton in order to connectthem to the cutting points The conductedexperiments confirm the accuracy of the factorsutilized and the results indicates a correctsegmentation at a rate of 97.2%.

FA7.5 P0188 1120 - 1140A 3D Reconstruction System based onImproved Spacetime StereoFederico TOMBARI, Luigi DI STEFANO, StefanoMATTOCCIA, Andrea MAINETTIUniversity of Bologna

Spacetime stereo is a promising technique foraccurate 3D reconstruction based on randomlyvarying illumination and temporal integration of thestereo matching cost. In this paper we show that thestandard spacetime stereo approach can be improvedin terms of accuracy of disparity estimation andconvergence speed by adoption of suitable matchingalgorithms based on adaptive support windows. We

also present a practical and cost-effective 3Dreconstruction system that deploys the proposedimproved spacetime method together with cheapcommercial off-the-shelf hardware (a PC, a stereocamera and a projector). Experimental results showthat the proposed system can yield rapidly accurate3D reconstruction of various types of objects andfaces.

104

Session FM1Date : Friday, 10 December 2010Time : 1310 - 1510Venue : Ball Room

Control Applications IChairs : Kirsten Moelgaard NIELSEN

Aalborg University

FM1.1 P0463 1310 - 1330Precision Motion Control for the ParallelMechanism of Virtual Axis Machine ToolGuoqin GAO, Yan LUO, Daogen JIANG, *Xinjun LIUJiangsu University*Tsinghua University

The precision motion control problem of a virtualaxis machine tool is addressed in this paper, which isstill one of the biggest obstacles of commercializationof the virtual axis machine tool. Firstly, thekinematics analysis of the parallel mechanism of a 6-DOF virtual axis machine tool is made. Secondly, ajoint-space trajectory planning based on the cycloidmovement is conducted to ensure the stability of themechanism movement. Finally, according to thecharacteristics of the parallel mechanism of thevirtual axis machine tool, a novel completelychattering-free sliding mode control system isdesigned to achieve the high-precision control of thevirtual axis machine tool. The simulation results showthat the control system has a good performance intracking and anti-interference and can achieve theaccurate control of the branches of the virtual axismachine tool. Furthermore, it is not necessary toestablish the accurate mathematical model of theparallel mechanism of the virtual axis machine tooland easy to design and implement.

FM1.2 P0258 1330 - 1350

Designing and Simulating of H∞ Controller forFlexible AircraftWen-Jie MENG, Ai-Jun LI, Jian TAN, Chang-QingWANGNorthwestern Polytechnical University

Reduction is essential in controller design forflexible aircraft, leading to model uncertainty andmodeling error. It is difficult to ensure robust stabilityby conventional controller. In order to resolve thatproblem, a robust controller, which can not onlycontrol flexible aircraft attitude but can also restrainstructure vibration quickly, is designed using mixedsensitivity control method. The controller is reducedto fourth-order and sixth-order by balancedtruncation. Controlling effects of the reductioncontroller are studied by simulating. Simulationresults show that compared with full-order controller,response performance of attitude angles using sixth-order controller is not affected and the controllingeffect for fourth-order structure transform mode ofaircraft is also roughly the same. Fourth-order

controller can make attitude angle of aircraft andvibration of structure transform mode stable and thestable time is similar to full-order controller.

FM1.3 P0824 1350 - 1410Temperature and Humidity Control in LivestockStablesMichael HANSEN, Palle ANDERSEN, Kirsten MoelgaardNIELSEN, Tom Soendergaard PEDERSEN, *MartinRIISGAARD-JENSENAalborg University*SKOV A/S

The paper describes temperature and humiditycontrol of a livestock stable. It is important to have acorrect air flow pattern in the livestock stable in orderto achieve proper temperature and humidity controlas well as to avoid draught. In the investigatedlivestock stable the air flow is controlled using wallmounted ventilation flaps. In the paper an algorithmfor air flow control is presented meeting the needsfor temperature and humidity while taking the airflow pattern in consideration. To obtain simple andrealisable controllers a model based control designmethod is applied. In the design dynamic models fortemperature and humidity are very importantelements and effort is put into deriving and testingthe models. It turns out that non-linearities aredominating in both models making feedbacklinearization the natural design method. The aircontroller as well as the temperature and humiditycontrollers are tested with promising results in a fullscale test stable.

FM1.4 P1024 1410 - 1430Backstepping Based Nested Multi-Loop ControlLaws for a QuadrotorErik DE VRIES, *Kamesh SUBBARAOTechnical University of Delft*The University of Texas at Arlington

Quadrotors are suitable testbeds for several pointto point hover based surveillance and observationtasks. As such, the hover mode is a very importantmode of operation for a quadrotor platform. Thispaper describes the design of a hover modecontroller using a backstepping based multi-loopdesign for a linearized quadrotor dynamics. Thiscontroller is then tested using both the linear and afull nonlinear simulation model of the quadrotor.

FM1.5 P1032 1430 - 1450Invariant Sets Techniques for Buck DC-DCConverter ControlAnamaria LUCA, Pedro RODRIGUEZ-AYERBE, DidierDUMUR, Pierre LEFRANCSUPELEC

This paper presents a simple and effective methodfor feedback law synthesis using invariant setstechniques. Given a time-varying Buck DC-DCconverter affected by perturbations and constraints,the proposed method gives the best feedback gain in

105

terms of finding the maximal invariant ellipsoidal set.This approach assures a good robustness but maylead to slow closed-loop performances and thereforea compromise will be made by imposing theplacement of the closed-loop poles in a certainregion. In order to reduce the invariant setconservativeness for the designed feedback gain, anew maximal invariant set is searched usingtruncated-ellipsoids.

FM1.6 P1064 1450 - 1510Nonlinear 3-D Trajectory Guidance forUnmanned Aerial VehiclesMousumi AHMED, Kamesh SUBBARAOThe University of Texas at Arlington

In this paper, we propose a backstepping basednonlinear guidance controller, designed for 3-D pathtracking for a UAV. The guidance law design assumesfirst order dynamics for the speed, heading andelevation angles. We demonstrate closed loopstability via Lyapunov analysis. The efficacy of thecontroller is demonstrated for two cases, namely thestraight path following and a curved path following.

Session FM2Date : Friday, 10 December 2010Time : 1310 - 1510Venue : Canary

Delay SystemsChairs : Pedro ALBERTOS

Universidad Polit´ecnica de ValenciaChengkang XIESouthwest University

FM2.1 P0229 1310 - 1330Sensor Fusion of Delay and Non-delay Signalusing Ensemble Kalman Filter with MovingCovarianceYucheng ZHOU, Jiahe XUChinese Academy of Forestry

This paper describes the design of ensembleKalman filter (EnKF) to implement fusion of the delayand non-delay data for nonlinear discrete-timesystem in order to achieve the excellent dynamicresponse. We proposed a fusion method with EnKFthat only needs to update the stored covariancebetween two different time instants, instead ofclassical method, which is re-performing Kalmanoperation at every step from the time of measureddelay signal to current time. To solve the fusionmethod, the EnKF algorithm is modified to obtainmembers of measurement ensemble fromuncorrelated sensors in the system but not a MonteCarlo method. With less computational costcomparing to the classical method and the uniformityof the computation in every iteration, the EnKF issuperior to extended Kalman filter (EKF) and offermuch advantage in terms of estimation performance,which is verified by using MATLAB simulation on thehigh-update rate Wheel Mobile Robot (WMR).

FM2.2 P0438 1330 - 1350Synchronization of Non-autonomous ChaoticSystems with Time-Varying Delay via DelayedFeedback ControlThongchai BOTMART, Piyapong NIAMSUP, *XinzhiLIUChiang Mai University*University of Waterloo

In this paper, we investigate the synchronizationof non-autonomous chaotic systems with time-varying delay via delayed feedback control. Using acombination of Riccati differential equation approach,Lyapunov-Krasovskii functional, inequality techniques,some new sufficient conditions for exponentiallystability of the error system are formulated in form ofa solution to the standard Riccati differentialequation. The designed controller ensures that thesynchronization of non-autonomous chaotic systemsare proposed via delayed feedback control. Numericalsimulations are presented to illustrate theeffectiveness of these synchronization criteria.

106

FM2.3 P0477 1350 - 1410Impulsive Exponential Stabilization of DiscretePopulation Growth Models with Time DelaysYu ZHANG, Jitao SUNTongji University

The purpose of this paper is to investigate theimpulsive exponential stabilization for the positiveequilibrium points of a class of discrete populationgrowth models with time delays. By using Lyapunovfunctionals, some new exponential stability criteriaare given. It is shown that impulses can indeed makeunstable equilibrium points exponentially stable, andwhen the impulses are employed to stabilize theunstable equilibrium points, the time interval betweenthe nearest two impulses should be small enough,i.e., impulses should act frequently. Two examplesare also presented to illustrate the effectiveness ofthe obtained results. It should be noted that, it's thefirst time that impulsive exponential stabilizationresults for discrete population models with timedelays have been given.

FM2.4 P0569 1410 - 1430Control of Unstable PDE with Long Input DelayYangyang ZHANG, Chengkang XIESouthwest University

It is well known that control of PDEs with inputdelay is just opening up for research. This paperpresents boundary control design and stabilization forareaction-diffusion PDE with diffusion coefficientsdepending on space variable and input time delay.The control design involves an interesting structure oftwo interconnected PDEs to transfer the input timedelay into a space extension. And the stabilityanalysisis characterized interms of the H1 norm ofthe actuator state rather than the usual L2.

FM2.5 P0651 1430 - 1450Robust Stabilization of 2-D State-delayedSystems with Stochastic PerturbationJuan YAO, Shuxia YE, Weiqun WANGNanjing University of Science and Technology

Delay systems may be encountered stochasticperturbation, and the stability conditions of suchsystems are different from delay systems andstochastic systems. This paper considers theproblems of robust stabilization for 2-D state-delayedsystem in Roesser model (RM) with stochasticperturbation for the first time. Delay-dependentsufficient stability conditions are derived in terms oflinear matrix inequalities (LMIs).A state feedbackcontroller is designed to ensure robust mean-squarestability of the closed-loop system. An example isprovided to demonstrate the effectiveness of theproposed approach.

FM2.6 P1131 1450 - 1510Decoupling MIMO Systems with Multiple

Input/Output Time DelaysPedro ALBERTOS, Pedro GARCIA GILUniversidad Polit´ecnica de Valencia

Decoupling multi-input multi-output systems is asimplifying technique to design, implement and tunecontrol systems. If there are also multiple delays, thedecoupling is even more interesting although there isalways a prize to pay and the resulting controllerscan become excessively complex. In this paper a newapproach to deal with these situations is presented.The procedure involves extracting non-delayedinformation to generate the control and canceling theinteractions by means of disturbance observers. Theapproach is general and it can be applied to unstableplants.

107

Session FM3Date : Friday, 10 December 2010Time : 1310 - 1450Venue : Kingfisher

Mobile Robotics IIIChairs : Sai Hong TANG

University Putra MalaysiaYongduan SONGBeijing Jiaotong University

FM3.1 P0421 1310 - 1330A Novel Approach to Motion Modeling usingFuzzy Cognitive Map and Artificial PotentialFieldsOmid Reza ESMAEILI MOTLAGH, Sai Hong TANG,Abdul Rahman RAMLI, Napsiah ISMAIL, D. NakhaeiNIAUniversity Putra Malaysia

Artificial potential field (APF) is well established forreactive navigation of robots. Initially, this paperdescribes a fast and robust fuzzy-APF on anActivMedia AmigoBot platform. Obstacle-relatedinformation is fuzzified just by sensory fusion whichresults in shorter runtime. The membership functionsof obstacles’ range and direction have been alsomerged into one function for smaller block of rules.The fuzzy-APF is tested and verified virtually withnon-concave obstacles. Main contribution of thisarticle is a new approach to motion modeling. Thegoal is to discover decision making behaviors of therobot in wayfinding. A novel decision modelingtechnique is developed based on capabilities of thefuzzy cognitive map (FCM) and supervised learningusing the genetic algorithm (GA). Decisionproductions for moving from one sub-space toanother are modeled in form of decision matrices.The robot trajectory under supply of such decisionmatrices has likelihood of nearly 90% with itstrajectory under the APF. Replication of robot motionis therefore achieved by modeling its decisionbehaviors in form of tangible matrices.

FM3.2 P0535 1330 - 1350Mobile Robot Localization using Feature BasedFuzzy Map MatchingHaoming XU, John James COLLINSUniversity of Limerick

This paper presents a novel scan matchingapproach entitled Fuzzy Map Matching for mobilerobot localization that extracts low level features inthe form of line segment from perceptual channelsthat are then matched to a map given a priori.Multiple candidate matches are supported throughthe use of fuzzy logic, which are iteratively refined.This probabilistic based fuzzy model of scan matchingis used to filter the alignment combination that havelow probability in order to reduce computationalcomplexity. In addition, the initial pose of the robot

does not have to be known as a result of support formultiple hypotheses with respect to potentialcorrespondences. Incomplete line segments thatresult from incomplete scans, noisy sensors, orocclusion do not present a problem as features inobservation space are grown during thecorrespondence phase of the algorithm. Thisapproach does not impose a heavy demand oncomputational resources, and is significantly lessresource hungry than the probabilistic approaches.Initial results demonstrated that the algorithmperforms well in real world environments.

FM3.3 P0660 1350 - 1410tinySLAM : A SLAM Algorithm in less than 200Lines C-Language ProgramBruno STEUX, Oussama EL HAMZAOUIMines ParisTech

This paper presents a Laser-SLAM algorithm whichcan be programmed in less than 200 lines C-languageprogram. The first idea aimed to develop andimplement a simple SLAM algorithm providing goodperformances without exceeding 200 lines in a C-language program. We use a robotic platform calledMinesRover, a six wheels robot with several sensors.We based our work and calculations on a laser sensorand the odometry of the robot. The article presentsthe different capabilities of the platform and the waywe use them in order to improve our programs. Wealso illustrates the difficulties encountered during theprogramming and testing of the algorithm. This workshows the possibility to perform complex tasks usingsimple and easily programmable algorithms.

FM3.4 P0680 1410 - 1430A DSP FPGA Based Hardware-in-the-LoopTesting PlatformTa-ming SHIH, *Ho-Chung CHANGMing Dao University*Chung Cheng Institute of Technology

Applying low-cost sensors for flight control is ofextreme challenge due to less accuracy nature ofsensors. Therefore a Hardware-in-the-Loop(HIL) ismost necessary for test and evaluation purpose toreduce the risk of flight testing. In this paper, a HILsystem is built up to test the autopilot hardwareperformance, control parameter tuning. By combiningthe commercial X-Plane fly simulation software andthe DSP and FPGA(Field Programmable Gate Array)based hardware interfaces, a 3D visualize HILplatform is built to improve test efficiency andreducing time and cost.

FM3.5 P1038 1430 - 1450A Fast Scan Matching for Grid-based LaserSLAM using Streaming SIMD ExtensionsOussama EL HAMZAOUI, Bruno STEUXMines ParisTech

The scan matching is one of the basic elements ofseveral SLAM (Simultaneous Localization and

108

Mapping) algorithms .There are a lot of researchesthat are interested in scan matching, but the majoritydeals with the algorithmic side only, without worryingabout implementation tricks that can be very useful.This paper presents a simple and effective method toaccelerate the scan matching step. Our method usesthe computing power offered by the SSE instructions.These instructions, developed by Intel, allows theprocessing of several data simultaneously. Thismethod does not require the use of a specifichardware (GPU for example), beceause SSEinstructions exist in the majority of the commonlyused processors (Intel, AMD). The examples andprograms presented in the article are part of ourSLAM algorithm, developed in the Robotics Center ofM nes ParisTech.

Session FM4Date : Friday, 10 December 2010Time : 1310 - 1510Venue : Nightingale

Factory Modeling and SimulationChairs : Dieter ZOEBEL

University Koblenz-LandauDaoping LIBeijing Institute of Technology

FM4.1 P0243 1310 - 1330Generic Data Types for Planning and Control inIntermodal Freight Transshipment ApplicationsDieter ZOEBEL, Christian WEYANDUniversity Koblenz-Landau

Planning and control environments for automatedguided vehicles are predominantly dedicated tospecialized vehicles, e.g. special carriers used inharbor areas for the transport of containers betweenquays and stackyards. The environment is completelydifferent for commercial standard vehicles whenoperated under automated guidance. Typically, theseare articulated vehicles with a high degree ofvariability, particularly with respect to their geometricand kinematic properties. Adaptivity requirementshave to be taken into account and lead to the designand development of data types with a high degree ofgenericity. How planning and control depend uponthose data types is explained in the broader scope ofa new intermodal role-on, roll-off traffic conceptwhich is currently under research. Furthermore, thedata types play a central role in the validationprocess to achieve the legal approval for theautomated guided traffic of standard commercialvehicles.

FM4.2 P0470 1330 - 1350Periodic Stable Robustness of SerialProduction Line with Single ProcessingParameter Perturbation based on Max-algebraDaoping LI, Xiao-Lan YAO, Qinghe WU, Xiao-DongZHANGBeijing Institute of Technology

The periodic stable robustness of serial productionline with single processing parameter is analyzedbased on max-algebra. A searching algorithm basedon golden section method to calculate the bound ofperturbation of processing parameter is alsoproposed. Example shows that the algorithm can getthe bound of the perturbation effectively, which ismeaningful to actual serial production line.

FM4.3 P0901 1350 - 1410Ontology-Based Product KnowledgeIntegration in Mobile EnvironmentCao LI, Zongwu XIE, Kui SUN, Hong LIU, *YongjunZHENG

109

Harbin Institute of Technology*Nottingham Trent University

This paper is to develop an ontology-based mobileenvironment for collaborative product design andmanufacture, which enables geographically dispersedteam members to collaborate over the internet withinthe total design or manufacturing process, includingproduct design specification, conceptual design, detaildesign , manufacture and recycling. In this paper, adatabase which based on a new method namedontology is built up to demonstrate the all relatedinformation in order to achieve an efficient andconvenient management. According these twoachievements, a new data management system isbuilt and tested in the mobile device and comparedto the traditional data management system at theend of this paper.

FM4.4 P0917 1410 - 1430Model Simplification for AcceleratingSimulation-based Evaluation of DispatchingRules in Wafer Fabrication FacilitiesTsung-Che CHIANGNational Taiwan Normal University

Dispatching rules are widely used for scheduling ofwafer fabrication facilities, and simulation is a populartool for evaluation of rules. A main weakness ofsimulation is the long computation time. In thispaper, we aim at simplifying the models of fabricationfacilities to reduce simulation time and meanwhilekeep the ability of evaluating rules correctly. Wepropose a metric to assess how well a simplifiedmodel evaluates rules and formulate the modelsimplification problem as a multiobjectiveoptimization problem. We also develop anevolutionary algorithm to solve the formulatedproblem. Experimental results show that simplifiedmodels generated by the proposed algorithm cansave half of simulation time and select high-performance rules.

FM4.5 P0620 1430 - 1450Binary Neural Network Classifier and it'sBound for the Number of Hidden LayerNeuronsNarendra S CHAUDHARI, *Aruna TIWARIIndian Institute of Technology, Indore*SGS Institute of Technology and Science, Indore

In this paper, a Binary Neural Network Classifier(BNNC) is proposed in which hidden layer training isdone in parallel. Learning Algorithm for the BNNC isdescribed, which is based on the principle of FastCovering Learning Algorithm (FCLA) proposed byWang and Chaudhari [1]. The BNNC offers highdegree of parallelism in hidden layer formation. Eachmodule in the hidden layer of BNNC is exposed to thepatterns of only one class. For achieving betteraccuracy, issue of overlapped classes are alsohandled. The method is tested on few benchmarkdatasets, accuracies are within the acceptable range.Due to parallelism at hidden layer level, training time

is decreased, therefore, it can be used for voluminousrealistic database. An analytical formulation isdeveloped to evaluate the number of hidden layerneurons, it is in the O(log(N)), where N representsthe number of inputs.

FM4.6 P0831 1450 - 1510Predicting Amount of Saleable Products usingNeural Network Metamodels of CasthousesAbbas KHOSRAVI, Saeid NAHAVANDI, DougCREIGHTON, Bruce GUNNDeakin University

This study aims at developing abstractmetamodels for approximating highly nonlinearrelationships within a metal casting plant. Metalcasting product quality nonlinearly depends on manycontrollable and uncontrollable factors. For improvingthe productivity of the system, it is vital for operationplanners to predict in advance the amount of highquality products. Neural networks metamodels aredeveloped and applied in this study for predicting theamount of saleable products. Training of metamodelsis done using the Levenberg-Marquardt and Bayesianlearning methods. Statistical measures are calculatedfor the developed metamodels over a grid of neuralnetwork structures. Demonstrated results indicatethat Bayesian-based neural network metamodelsoutperform the Levenberg-Marquardt-basedmetamodels in terms of both prediction accuracy androbustness to the metamodel complexity. In contrast,the latter metamodels are computationally lessexpensive and generate the results more quickly.

110

Session FM5 - Invited SessionDate : Friday, 10 December 2010Time : 1310 - 1450Venue : Pelican

Human Robot InteractionChairs : Sarath KODAGODA

University of Technology, SydneyRajesh Elara MOHANSingapore Polytechnic

FM5.1 P0842 1310 - 1330Experimenting Extended Neglect ToleranceModel for Human Robot Interactions in ServiceMissionsRajesh Elara MOHAN, *Carlos A ACOSTA CALDERON,Changjiu ZHOU, **Wijerupage Sardha WIJESOMASingapore Polytechnic*Singapore polytechnic**Nanyang Technological University

In this paper, we validate the extended neglecttolerance model for estimation of human robot teamperformance in relation to robot autonomy andcompare its results with the traditionally adoptedneglect tolerance model which assumes zero falsealarms in human robot interactions. Extended neglecttolerance model estimates robot performance inhuman robot teams, where the human operatorswitches control between robots sequentially, basedon acceptable performance levels, taking into accountany false alarms and their respective demands.Experiments were performed with Robo-Erectus@Home, a service robot across tele-operation,and semi-autonomous modes of autonomy where ahuman operator controlled the robot to perform awalking assistant task. Measured false alarmdemands and robot performances were largelyconsistent with the extended neglect tolerance modelpredictions for both autonomy modes. We alsocompared traditionally adopted neglect tolerance andextended neglect tolerance model for the sameexperimental design. The results showed that thelatter offers better estimates of robot performanceand attention demands, due to the inclusion of falsealarms into the model.

FM5.2 P0981 1330 - 1350Modelling of Human Motor Control in anUnstable Task through Operational SpaceFormulationShou-Han ZHOU, Denny OETOMO, Iven MAREELS,*Etienne BURDETThe University of Melbourne*Imperial College, London

Human motor control computational model is animportant component in the study and the successfulrealisation of human-robot interaction. In this paper,the Operational Space Formulation is presented as asuitable framework of human motor control

computational model based on the Equilibrium PointHypothesis (EPH) approach. The iterative adaptivecontrol strategy was incorporated to simulate humanmotor adaptation to different tasks. The strategyinvolves the use of an Equilibrium Model whichrepresents the ideal human motor response to agiven task. The combined strategy was simulated tomatch a set of data gathered experimentally fromseveral human subjects. The results were observedto explain many of the features found in the recordedbehaviours in the EPH-based approach of humanmotor modelling.

FM5.3 P0993 1350 - 1410Using Common Motion Patterns to Improve aRobot's Operation in Populated EnvironmentsStephan SEHESTEDT, Sarath KODAGODA, GaminiDISSANAYAKEUniversity of Technology, Sydney

Robotic devices are increasingly penetrating thehuman work spaces as stand alone units and helpers.It is believed that a robot could be easily integratedwith humans, if the robot can learn how to behave ina socially acceptable manner. This involves a robot toobserve, learn and comply with basic rules of humanbehaviors. As an example, one would expect a robotto travel in an environment without intruding humanworkspaces unnecessarily. Thus, identifying commonmotion patterns of people by observing a specificenvironment is an important task as people'strajectories are usually not random, however aretailored to the way the environment is structured. Wepropose a learning algorithm to construct a SampledHidden Markov Model (SHMM) that captures behaviorof people through observations and thendemonstrate how this model could be exploited forplanning socially aware paths. Experimental resultsare presented to demonstrate the viability of theproposed approach.

FM5.4 P1042 1410 - 1430Thinking Head: Towards Human CentredRoboticsDamith C HERATH, Christian KROOS, Catherine JSTEVENS, *Lawrence CAVEDON, **PrashanPREMARATNEUniversity of Western Sydney*RMIT University**University of Wollongong

Thinking Head project is a multidisciplinaryapproach to building intelligent agents for humanmachine interaction. The Thinking Head Frameworkevolved out of the Thinking Head Project and itfacilitates loose coupling between variouscomponents and forms the central nerve system in amultimodal perception-action system. The paperpresents the overall architecture, components andthe attention system. The paper then concludes witha preliminary behavioral experiment that studies theintelligibility of the audiovisual speech outputproduced by the Embodied Conversational Agent

111

(ECA) that is part of the system. These resultsprovide the baseline for future evaluations of thesystem as the project progresses through multipleevaluate and refine cycles.

FM5.5 P0759 1430 - 1450Fuzzy Direct Adaptive Control for RoboticManipulatorsDaoxiang GAO, Qingchun WANG, Dunmin LUBeijing Forestry University

A fuzzy direct adaptive control strategy isproposed for the trajectory tracking control of roboticmanipulators. Different to the traditional fuzzyadaptive control method, a generalized hyperbolicmodel is used to be the fuzzy basis function. Theinput generalized variables of the fuzzy logic system,derived from the input variables after theirtransformation, can cover with the whole input space.Thus, the fuzzy logic system can approximate thedesired control input by an arbitrary accuracy.Furthermore, due to the special structure of the fuzzyhyperbolic model, the adaptive fuzzy control strategyneeds only fewer updated parameters so that it canavoid a heavy computational burden but getting asatisfied tracking performance. Simulation studies arepresented to show that this control algorithm has anexpected tracking accuracy.

Session FM6Date : Friday, 10 December 2010Time : 1310 - 1510Venue : Oriole

Tracking and SurveillanceChairs : Geoff A WEST

Curtin University of Technology

FM6.1 P0702 1310 - 1330Head Plane Estimation Improves the Accuracyof Pedestrian Tracking in Dense CrowdsIrshad ALI, Matthew N DAILEYAsian Institute of Technology

Human detection and tracking in high densitycrowds is an unsolved problem. Standardpreprocessing techniques such as backgroundmodeling fail when most of the scene is in motion.Because of high levels of occlusion, dense features,and shadows, object detectors tend to produce largenumbers of false detections. We introduce a newmethod based on 3D head plane estimation thatreduces these false detections while preserving highdetection rates. Our algorithm learns the head planefrom observations of human heads, without any apriori extrinsic camera calibration information. In anexperimental evaluation, we show that the headplane estimation technique dramatically improves theperformance of a pedestrian tracker for dense crowdsbased on a Viola and Jones AdaBoost cascadeclassifier for head detection, a particle filter fortracking, and color histograms for appearancemodeling.

FM6.2 P0946 1330 - 1350Robust Object Tracking using Local OrientedEnergy Features and its Hardware/SoftwareImplementationEhsan NOROUZNEZHAD, Abbas BIGDELI, AdamPOSTULA, *Brian C LOVELLNational ICT Australia*The University of Queensland

This paper presents the use of local orientedenergy features for real-time object tracking onembedded vision systems. Local oriented energyfeatures are extracted using complex Gabor filters.Filtering is carried out across multiple channels withdifferent frequencies and orientations. Theeffectiveness of the chosen feature set is tested usinga mean-shift tracker. Our experiments show thatadding local oriented energy features can significantlyenhance the performance of the tracker in presenceof photometric variations and geometrictransformation. The realtime implementation of thesystem is also described in this paper. To achieve thedesired performance, a hardware/software codesignapproach is pursued. Multi-channel Gabor filtering,Local Oriented Energy Feature and Feature histogramComputation is implemented on hardware whilemean-shift vector calculation is performed on a

112

processor. The system was synthesized onto a XilinxVirtex-5 XC5VSX50T using Xilinx ML506 developmentboard and the implementation results are presented.

FM6.3 P0844 1350 - 1410Omnidirectional Video Stabilisation on a VirtualCamera using Sensor FusionThomas ALBRECHT, Tele TAN, Geoff A WEST,*Thanh LYCurtin University of Technology*Defence Science and Technology Organisation

This paper presents a method for robustlystabilising omnidirectional video given the presenceof significant rotations and translations by creating avirtual camera and using a combination of sensorfusion and scene tracking. Real time rotationalmovements of the camera are measured by anInertial Measurement Unit (IMU), which provides aninitial estimate of the ego-motion of the cameraplatform. Image registration is then used to refinethese estimates. The calculated ego-motion is thenused to adjust an extract of the omnidirectionalvideo, forming a virtual camera that is focused on thescene. Experiments show the technique is effectiveunder challenging ego-motions and overcomesdeficiencies that are associated with unimodalapproaches making it robust and suitable to be usedin many surveillance applications.

FM6.4 P0852 1410 - 1430Square Patch Feature: Faster Weak-Classifierfor Robust Object DetectionYasir MOHD MUSTAFAH, *Abbas BIGDELI, AmeliaWong AZMAN, *Farhad DADGOSTAR, Brian C LOVELLThe University of Queensland*National ICT Australia

This paper presents a novel generic weak classifierfor object detection called “Square Patch Feature”.The speed and overall performance of a detectorutilizing Square Patch features in comparison to otherweak classifiers shows improvement. Each weakclassifier is based on the difference between two orfour fixed size square patches in an image. A pre-calculated representation of the image called “patchimage” is required to accelerate the weak classifierscomputation. The computation requires fewerarithmetic operations and fewer accesses to the mainmemory in comparison to the well known Viola-JonesHaar-like classifier. In addition to the fastercomputation, the weak classifier can be extended forin-plane rotation, where each square patch can berotated to detect in-plane rotated objects. The resultsof the experiments on the MIT CBCL Face datasetshow that a Square Patch Feature classifier is asaccurate as the Viola-Jones Haarlike classifier, andwhen implemented on hardware (i.e. FPGA), it isalmost 2 times faster.

FM6.5 P0329 1430 - 1450A Hybrid Computational Model for Spoken

Language UnderstandingGuangpu HUANG, Meng Joo ERNanyang Technological University

This paper shows that the integration of statisticaland connectionist methods can greatly enhancehuman-computer interaction through speech. Theresearch approach is inspired by recent advances inhigh performance automatic speech recognition(ASR)systems and neurocognitive researches of naturallanguage understanding(NLU). And a modest hybridcomputational model is proposed and implemented toachieve intelligent spoken language understanding(SLU) in an information retrieval system.

FM6.6 P1142 1450 - 1510Frontal Face Classifier using AdaBoost withMCT FeaturesJongmin YOON, Daijin KIMPOSTECH

In this paper, we describe how to classify frontalface from the results of face detection which includenon-frontal faces. To do this, we use AdaBoostlearning method with Modified Census Transform(MCT) to construct a two-class classifier. As a resultof that, our frontal face classifier achieves highclassification rate above 96% and fast performanceabout 10 frames/sec in mobile device.

113

Session FM7Date : Friday, 10 December 2010Time : 1310 - 1510Venue : Bluebird

Pattern RecognitionChairs : Sundararajan NARASIMHAN

Nanyang Technological University

FM7.1 P0159 1310 - 13303D Object Recognition via Multi-ViewInspection in Unknown EnvironmentsJamie WESTELL, Parvaneh SAEEDISimon Fraser University

This paper presents a system for objectrecognition and localization within unknown indoorenvironments. The system includes a GUI designthrough which the user may describe an object ofinterest by means of color, size, and shape. A novelcoarse to fine identification mechanism thatincorporates multiple views of an object is then usedto locate the described object within an unknownenvironment. The system includes a training stage inwhich representative information is extracted fromdatabase images. A stereo vision system, mountedon an indoor robot platform (Fig. 1), is used toretrieve the 3D location of potential match candidatesin the scene and to inspect possible matches fromthree distinct viewpoints. Experimental evaluation isperformed for indoor environments and promisingresults are shown for the application of this system.

FM7.2 P0571 1330 - 1350Large Object Detection in ClutteredBackground using Boosted Markov ChainMonte CarloSungho KIM, *Jungho KIM, *Chaehoon PARK, *In SoKWEONYeungnam University*Korea Advanced Institute of Science and Technology

In this paper, we present a new object detectionmethod using codebook and boosted Markov ChainMonte Carlo (MCMC) estimation. It is relatively welldetected using adaboost and simple Haar-likefeatures for small objects. However, the detectionproblem is more difficult when object size becomeslarger (over 150 x 150) due to different surfacemarkings and clutter. Codebook-based objectrepresentation and boosted MCMC method can detectlarge objects robustly. Experimental results validateconvincing detection for large objects.

FM7.3 P0592 1350 - 1410Exploiting Feature Dynamics for Active ObjectRecognitionPhilipp ROBBEL, Deb ROYMassachusetts Institute of Technology

This paper describes a new approach to object

recognition for active vision systems that integratesinformation across multiple observations of an object.The approach exploits the order relationship betweensuccessive frames to derive a classifier based on thecharacteristic motion of local features across visualsweeps. This motion model reveals structuralinformation about the object that can be exploited forrecognition. The main contribution of this paper is arecognition system that extends invariant localfeatures (shape contexts) into the time domain byintegration of a motion model. Evaluations on onestandardized and one custom collected dataset fromthe humanoid robot in our laboratory demonstratethat the motion model allows higher-qualityhypotheses about object categories quicker than abaseline system that treats object views as unorderedstreams of images.

FM7.4 P0665 1410 - 1430Multiple Kernel Learning SVM-based EMGPattern Classification for Lower Limb ControlQingshan SHE, Zhizeng LUO, Ming MENG, Ping XUHangzhou Dianzi University

Based on multiple kernel learning (MKL) supportvector machine and decision tree combined strategy,a multi-class classification method is proposed toclassify lower limb motions using electromyography(EMG) signals. According to the framework ofmultiple kernel learning, the MKL-based multi-classifier is constructed using binary treedecomposition method. Four-channel surface EMGsignals are firstly collected from lower limb muscles,and then some time-domain features are extractedand inputted into the proposed multi-classifier. Fivesubdividing patterns are finally identified in levelwalking, i.e. support prophase, support metaphase,support telophase, swing prophase and swingtelophase. The experimental results show that theproposed method can successfully identify thesesubdividing patterns with better accuracy thanstandard single-kernel support vector machineclassifier.

FM7.5 P0817 1430 - 1450Classification of Abnormalities in DigitizedMammograms using Extreme LearningMachineVani GOPALAN, Savitha RAMASAMY, SundararajanNARASIMHANNanyang Technological University

Digital mammography is a preferred method forearly detection of breast cancer. However, in mostcases, it is very difficult to distinguish benign andmalignant masses without a biopsy, hence,misdiagnosis is always possible. In this paper, theExtreme Learning Machine (ELM) algorithm is used toclassify the suspicious masses in digitizedmammograms available in the Mini-MIAS database.As selection of features is critical in efficientclassification of mammograms, a study wasconducted to identify the best features that make a

114

clear distinction between the malignant and benignmasses. Then, the recently developed ExtremeLearning Machine (ELM) based classifier is used toclassify the benign and malignant cases. It isobserved that the performance of the batch learningELM algorithm is dependent on the number of hiddenneurons, and the magnitude of input weightinitialization. An extensive study is conducted toidentify the best number of neurons to achieve thebest training and testing classification performance.The performance results are then compared with theclassification results, available in the literature. Theperformance study results show that the classificationefficiency of the ELM classifier is better than theother existing algorithms, for the mammogramclassification problem of the database considered forthis study.

FM7.6 P0945 1450 - 1510A Divide-and-Distribute Approach to Single-Cycle Learning HGN Network for PatternRecognitionAnang Hudaya MOHAMMAD AMIN, Asad I. KHANMonash University, Melbourne

Distributed Hierarchical Graph Neuron (DHGN) is asingle-cycle learning distributed pattern recognitionalgorithm, which reduces the computationalcomplexity of existing pattern recognition algorithmsby distributing the recognition process into smallerclusters. This paper investigates an effect of dividingand distributing simple pattern recognition processeswithin a computational network. Our approachextends the single-cycle pattern recognition capabilityof Hierarchical Graph Neuron (HGN) for wirelesssensor networks into the more generic framework ofcomputational grids. The computational complexity ofthe hierarchical pattern recognition scheme issignificantly reduced and the accuracy is improved.The single-cycle learning capability, which developswithin the HGN, shows better noisy patternrecognition accuracy when size of the clusters isadapted to pattern data. The scheme lowers storagecapacity requirements per node and incurs lessercommunication complexity while retaining HGN’sresponse-time characteristics. Higher recall accuracyand scalability of the scheme is tested by storinglarge numbers of binary character patterns andheterogeneous binary images. The results show thatthe response-time remains insensitive to the numberof stored pattern, the accuracy is improved, and thesystem resource requirements are significantlyreduced.

Session FP1 - Invited SessionDate : Friday, 10 December 2010Time : 1600 - 1720Venue : Ball Room

Underwater RoboticsChairs : Norimitsu SAKAGAMI

Tokai University

FP1.1 P0959 1600 - 1620Development of the Actuator ConcentrationType Removable Underwater ManipulatorFumiaki TAKEMURA, Reyes Tatsuru SHIROKUOkinawa National College of Technology

Authors have been developing the actuatorconcentration type removable underwatermanipulator. This manipulator has the followingfeatures. 1. This manipulator makes it relatively easyto waterproof due to actuator concentration. 2.Wireless LAN Access Point(AP) is installed inunderwater robot and AP connects with operating PCby wire LAN. This manipulator has a few cameras totake a look operation. The command to each actuatorof manipulator and camera image communicate withoperating PC via wireless LAN. Therefore, theexternal signal wire for control has no use. 3.Removable is easy because this manipulator isconnected with the underwater robot body only withthe power cable. In this paper, we verify thefeasibility of 2 by wireless LAN communicationexperiment using seawater. Moreover, we describethe design of manufactured actuator concentrationtype removable manipulator and the experimentalresult.

FP1.2 P0967 1620 - 1640Adaptive Region Tracking Control forAutonomous Underwater VehicleXiang LI, Saing Paul HOU, Chien Chern CHEAHNanyang Technological University

This paper presents an adaptive region trackingcontrol for Autonomous Underwater Vehicle (AUV).The AUV is required to track a moving region toaccomplish a given task. The desired target isspecified as a region rather than a point so that thecontrol effort used to track the region is minimal. Inthe applications where the accuracy is of utmostimportance, the desired region can be chosen to besmall so that the precision is not lost. The desiredregion can be scaled up or scaled down so that theAUV can adjust its position to suit the applications. ALyapunov-like function is presented for the stabilityanalysis. Simulation results on AUV with 6 degrees offreedom are presented to demonstrate theeffectiveness of the proposed controller.

FP1.3 P1000 1640 - 1700A Neuro-Fuzzy Controller for Underwater

115

Robot ManipulatorsShunmugham R PANDIAN, *Norimitsu SAKAGAMISoutheastern Louisiana University*Tokai University

Autonomous underwater vehicles are increasinglyreplacing the prevalent remotely operated vehicle-manipulator systems. Most current generation AUVsare not fitted with manipulators and hence are mainlylimited to underwater surveying and surveillancetasks because of the difficulty in the coordinatedcontrol of the resulting underwater vehicle-manipulator systems. While several researchers haveproposed various techniques for control of AUVs,there is still much research to be done on the precisecontrol of underwater manipulators. This paperpresents an intelligent control method for underwatermanipulators based on the neuro-fuzzy approach.The controller is composed of fuzzy PD control withfeedback gain tuning by linguistic rules. A neuralnetwork compensator approximates the dynamics ofthe multiple degrees of freedom manipulator indecentralized form. The proposed controller hasadvantages of simplicity of implementation due todecentralized design, precision, and robustness topayload variations and hydrodynamic disturbances. Ithas lower energy consumption compared to theconventional PD control method. The effectiveness ofthe proposed controller is illustrated by experimentalresults for a three degrees of freedom underwatermanipulator.

FP1.4 P1036 1700 - 1720Design and Development of an Attitude ControlSystem for a Human-sized ROVNorimitsu SAKAGAMI, *Takafumi KANAYAMA,**Tomohiro UEDA, **Hideki HASHIZUME, **MizuhoSHIBATA, ***Hiroyuki ONISHI, ***ShigeoMURAKAMI, **Sadao KAWAMURATokai University*DENSHI KOGYO CO., LTD.**Ritsumeikan University***DAINIPPON SCREEN MFG. CO., LTD.

In this paper, we present the design anddevelopment of an attitude control system for ahuman-sized ROV (Remotely Operated Vehicle) witha dual-manipulator system. The main operations ofour ROV are sampling operations for biologicalresearches, geological researches and archaeologicalexplorations in Lake Biwa, the biggest lake in Japan.In order to achieve these operations, we propose anattitude control system and explain the availability ofthe proposed system in this paper. For the design ofa prototype of the attitude control system, simulationanalysis was conducted. Based on the analysis, apitch angle control system was designed anddeveloped as the prototype of the attitude controlsystem. An experiment was conducted in a test tankto demonstrate the effectiveness of the proposedsystem. The result shows that the developed systemis useful for pitch angle control of the ROV.

Session FP2Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Canary

Feature Extraction, Grouping andSegmentation IIChairs : Tele TAN

Curtin University of TechnologyFrederic MAIREQueensland University of Technology

FP2.1 P0212 1600 - 1620A Unified 2D-3D Video Scene Change DetectionFramework for Mobile Camera PlatformsWilson Suryajaya LEOPUTRA, Tele TAN, SvethaVENKATESHCurtin University of Technology

In this paper, we present a novel scene changedetection algorithm for mobile camera platforms. Ourapproach integrates sparse 3D scene backgroundmodelling and dense 2D image background modellinginto a unified framework. The 3D scene backgroundmodelling identifies inconsistent clusters over time ina set of 3D cloud points as the scene changes. The2D image background modelling further confirms thescene changes by finding inconsistent appearances ina set of aligned images using the classical MRFbackground subtraction technique. We evaluate theperformance of our proposed system on a number ofchallenging video datasets obtained from a cameraplaced on a moving vehicle and the experimentsshow that our proposed method outperformsprevious works in scene change detection, whichsuggested the feasibility of our approach.

FP2.2 P0431 1620 - 1640Flying Vision System of Placement Machine forMEMS MicrophoneLiguo CHEN, Jifeng ZHU, Weibin RONG, Lining SUNHarbin Institute of Technology

In the die bonding process of MEMS microphonechip, the exact center and rotate angle of MEMSmicrophone chip is corrected after the vacuum nozzlehas adsorbed it. In order to achieve high speed andprecision placement function, the flying vision systembased on rotating mirror is proposed, the mirror isdriven to rotate by the power drawn from the nozzlerise and fall movement through a rack and pinionmechanism, the nozzle pick and place functions is notinterfered by mirror through designing the rack andpinion mechanism parameters. The vision system cancapture an static clear MEMS chip image when themirror rotated to a 45 degrees position referencedwith horizontal position, the center position androtating angle of MEMS chip is fast calculated by thefine and coarse Hough Transform Algorithm, thecenter location and angle will be corrected by thecontrol system during the mounted head is flying to

116

placement position. The experimental results showthat the system can obtain high-quality MEMSmicrophone chip image, the chip’s center positionand rotating angle correction algorithm takes nomore than 0.1s, so it can meet the high-speedplacement requirements. In this proposed flyingvision system, mechanical structure is simple andcompact, MEMS chip’s image is clear withoutdistortion, and there is no restriction on the mountedhead movement trajectory. The flying vision systemsignificantly enhances the quality and efficiency ofSMD Chip; it can be widely used in the surfacemounted system.

FP2.3 P0623 1640 - 1700Improve Library Shelf Reading using ColorFeature Matching of Book-Spine ImagesSpencer G FOWERS, Dah-Jye LEE, *GuangmingXIONGBrigham Young University*Beijing Institute of Technology

Many computer vision algorithms for patternmatching, object tracking, and 3-D reconstruction,etc., begin with feature detection and matching.Common feature detectors such as Harris, Sobel,Canny, and Difference of Gaussians perform basiclinear algebra operations on an image in order toidentify "corners" or "edges" for matching. Thesedetectors however, require single-channel (grayscale)source images. For color source images, colorchannels are typically averaged or converted tocreate grayscale images for processing, discarding alarge amount of highly useful information. This paperoutlines the proposed color Difference of Gaussians(DoG) algorithm for feature detection, whichoutperforms the grayscale DoG in number and qualityof features found. The new color DoG is applied to acolor feature matching application for improving thelibrary inventory (shelf-reading) process.Experimental results demonstrate the robustness ofthis color feature detector.

FP2.4 P0681 1700 - 1720A Fast and Robust Descriptor for Multiple-viewObject RecognitionMaja RUDINAC, Pieter P JONKERDelft University of Technology

In this paper we propose a fast and robustdescriptor for multiple view object recognition using asmall number of training examples. In order to designa descriptor to be discriminative between manydifferent object appearances, we base it on acombination of invariant color, edge and texturedescriptors. We use a color descriptor based on aHSV histogram - as it is robust to size and position ofthe object -, a gray level co-occurrence matrix astexture descriptor and an edge histogram as shapedescriptor. After extraction of feature vectors, weperform normalization on all feature vectors from thetraining database in order to increase the importanceof the most dominant feature components and

reduce the less dominant ones. This normalizationimproves the recognition performance with almost30% in case of a small number of training objectsand in case of noise or occlusion. We tested ourdescriptor on the Columbia Object Image Librarydataset (COIL 100) which presents objects in scaled,translated and rotated versions. Our recognition rateis extremely high: 99% in case of a large number oftraining objects and 93% for training with only 4views of the object, or 5% of the database. Thedescriptor was also tested under various distortions:illumination change, noise corruption and occlusions.It proved to be very robust, with recognition ratesdecreasing only less then 5%. We compared ourresults with state of the art methods and weconclude that our descriptor achieves a betterperformance, both on the regular COIL database andon all distorted variants.

FP2.5 P1090 1720 - 1740Obstacle-Free Range Determination for RailTrack Maintenance VehiclesFrederic MAIRE, *Abbas BIGDELIQueensland University of Technology*National ICT Australia

Maintenance trains travel in convoy. In Australia,only the first train of the convoy pays attention to thetrack signalization (the other convoy vehicles simplyfollow the preceding vehicle). Because of humanerrors, collisions can happen between themaintenance vehicles. Although an anti-collisionsystem based on a laser distance meter is already inoperation, the existing system has a limited rangedue to the curvature of the tracks. In this paper, weintroduce an anti-collision system based on vision.The two main ideas are, (1) to warp the cameraimage into an image where the rails are parallelthrough a projective transform, and (2) to track thetwo rail curves simultaneously by evaluating smallparallel segments. The performance of the system isdemonstrated on an image dataset.

FP2.6 P0659 1740 - 1800Recognizing Human Motions from SurroundingViewpoints Employing HierarchicalEigenspacesS. M. ASHIK EFTAKHAR, Joo Kooi TAN, HyoungseopKIM, Seiji ISHIKAWAKyushu Institute of Technology

The development of an automatic human motionrecognition system leads to the solution to theproblems concerning the video-based applications inrecognizing human activities. Such a system is to beinvestigated in the context of human motion analysis.Although there were a large number of researches inthis area for a long time, there was little attentiongiven to the development of a structured databasefor successful retrieval of motion data incorporatingthe time-space trade-off. We have proposed a systemwhich is capable of dealing with large set of motiondata employing an efficient database structure with

117

improved performance. We have analyzed twomotion representation techniques to realize theeffectiveness of the system. Performance evaluationis performed by synthesized 3D human motionsobserved from eight camera directions. Finally, ourresults show that the proposed recognition schemeperforms well for the captured motions.

Session FP3Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Kingfisher

Neuro-Fuzzy Control IIChairs : Meng Joo ER

Nanyang Technological UniversityXiang LISingapore Institute of ManufacturingTechnology

FP3.1 P0983 1600 - 1620An Intelligent Control Approach for BloodPressure System using Self-generating FuzzyNeural NetworksFan LIU, Meng Joo ERNanyang Technological University

This paper presents an intelligent control approachfor blood pressure system using self-generating fuzzyneural networks (SGFNN). The proposed SGFNN issimple and effective and is able to generate a fuzzyneural network to model unknown nonlinearities ofcomplex blood pressure system. This paperinvestigates the use of fuzzy neural networktechnique for modeling and automatic control ofmean arterial pressure (MAP) through theintravenous infusion of sodium nitroprusside (SNP).Simulation studies based on a sensitive model of MAPillustrate the ability of the proposed approach inmodeling drug delivery system and controlpostsurgical blood pressure.

FP3.2 P0455 1620 - 1640Neural Network Based Minimal State-SpaceRepresentation of Nonlinear MIMO Systems forFeedback ControlKristina VASSILJEVA, Eduard PETLENKOV, JuriBELIKOVTallinn University of Technology

A state-space technique for control of nonlinearmulti-input multi-output (MIMO) systems identified byan Additive Nonlinear Autoregressive eXogenous(ANARX) model is presented. Controlled system isidentified by Neural Network based SimplifiedAdditive NARX (NN-SANARX) model linearized bydynamic feedback. The neural network based modelis represented in the discrete-time state-space form.The problem of finding the minimal state-spacerepresentation is considered.

FP3.3 P1008 1640 - 1700Metrics for Measurement of Additive Noise toWeight in Sigmoidal FFANNsAmit Prakash SINGH, Chandra Shekhar RAI, *PravinCHANDRAGGS Indraprastha University*University of Delhi South Campus

118

Feedforward artificial neural networks aresusceptible to thermal noise when implemented asanalog or digital hardware. Thus to study of theeffect of noise on neural hardware, noise metricshave to be identified. In this paper, five new metricsare prepared and shown to be relevant for noisemodeling in the neural hardware.

FP3.4 P0534 1700 - 1720Fuzzy Metering Control on the Auxiliary RoadSignal of the Expressway ExitChangliang YUAN, Honghai LIResearch Institute of Highway Ministry of Transport

This paper gives the basic description of fuzzymetering control of the auxiliary road signals. On thebase of this description, the basic fuzzy logic controlstructure and fuzzy control algorithm are described.The necessity of using the exit-auxiliary-road densityas the fuzzy metering control objective is analyzed.The real membership function is used in the designprocess of the auxiliary road fuzzy controller. TheXueYuan Road exit on the second ring road in Beijingis selected as the study object in the simulation, andthe simulation results reflect the adjustment resultsof the fuzzy controller with the present fuzzy controlrules. The metering rate only changes a little.However, when the average delay is selected as theevaluation parameter, the fuzzy metering control willproduce less delay than the field cycle-split controlstrategy. The exit delay of the expressway decreases66%. The delay of the auxiliary road downstream ofthe stop line decreases 93%.

FP3.5 P0743 1720 - 1740Designing a Fuzzy Logic Controller to Adjustthe Angle of Tires in Four Wheel SteeringVehiclesAli HAKIMA, Sina AMELIQazvin Islamic Azad University

In this paper, a 4 wheel steering vehicle (4WS)was used as a template and a vehicle was modeled(experimental) by using 2 degree of freedom system.The dynamics governing the Yaw behavior of avehicle has been simulated with MATLAB/SIMULINKsoftware. A controller, working by fuzzy logic wasthen designed. A relationship between front and rearangle of tires in different speeds and different frontsteering angle was found after running simulation forover 600 times. This controller increases thedirectional stability property of experimental vehiclewhen this vehicle is turning or maneuvering. Therelationship was then used to make rules for thefuzzy logic controller. Subsequently, the experimentalvehicle was compared with an uncontrolled vehicleand a vehicle with a very high lateral stiffness whichthere is infinitesimal slipped on turning ormaneuvering (reference). This allows us to correctlyidentify the amount of slipping in our experimentalvehicle. Our results show significant improvement ofdirectional stability property of experimental vehiclein comparison with uncontrolled vehicle in different

speed and different steering angles. Also the gottenresults show that the side slip angle and the yaw rateof experimental vehicle were enhanced significantly.

FP3.6 P0447 1740 - 1800An Enhanced Online Self-organizing FuzzyNeural NetworkLinn SAN, Meng Joo ER, *Xiang LI, Lainyin ZHAI,Amin TORABI JAHROMINanyang Technological University*Singapore Institute of Manufacturing Technology

An Enhanced Online Self-organizing Fuzzy NeuralNetwork (EOS-FNN) is proposed in this paper. Theproposed algorithm can improve computationalefficiency while achieving comparable performanceand accuracy compared to other methods. Theproposed EOS-FNN starts with an empty rule set andautomatically generates fuzzy rules according to theproposed criteria during the learning process. All theparameters of the fuzzy rules are updated by theExtended Kalman Filter (EKF) method. Nonlineartime-series prediction processes are used to evaluatethe performance of the proposed EOS-FNN algorithmwith a comparison to other popular algorithmsincluding DFNN, GDFNN and FAOS-PFNN. Simulationresults have shown that the proposed algorithmreduces computation time while achievingcomparable accuracy.

119

Session FP4Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Nightingale

Control Applications IIChairs : Arun D MAHINDRAKAR

Indian Institute of Technology,Madras

FP4.1 P0833 1600 - 1620Improvement of Short-Span SeekingPerformance in a Dual-Stage Hard Disk DriveActuatorBranislav HREDZAKThe University of New South Wales

This paper proposes an alternative method toimprove short-span seeking performance of acompensator based control system for a dual-stageservo actuator. The short-span seeking performanceis improved by presetting initial conditions of thecompensator during the short-span seek commands.This results in faster response of the secondarycontrol loop and hence in faster seek times within therange of the secondary actuator. The performance ofthe proposed initial conditions based method iscompared with the performance of recently proposedmethods using constant and dynamic saturations inthe secondary loop. The results show that theperformance of the proposed method is comparablewith the performance of the method using thedynamic saturation and thus can be used as analternative, simpler method, to achieve faster short-span seek times.

FP4.2 P0146 1620 - 1640Stabilization of a Circular Ball-and-BeamSystemAakash A RAVICHANDRAN, Arun D MAHINDRAKARIndian Institute of Technology, Madras

The ball on a circular beam system is anunderactuated mechanical system that is challengingcontrol problem compared to its variant: the ball on ahorizontal beam. The control challenge stems outfrom the fact that the ball on a circular beam systemhas two unstable eigenvalues unlike its variant whichhas one. A stabilizing controller synthesized usingbackstepping methodology is presented. In doing so,a series of nonlinear coordinate transformations isapplied on the equations of motion to arrive at astate-space representation which is in the so calledquadratic non-triangular normal form. Simulationresults are presented to demonstrate theeffectiveness of the proposed controller in stabilizingthe system while respecting the physical constraintson the system.

FP4.3 P0471 1640 - 1700Efficient Control System Development usingReal-time Virtual Hardware-in-the-loopSimulationJun ZHONG, Yi JIN, Guofu LIAN, Guoliang DING,Changan ZHUUniversity of Science and Technology of China

In this paper, an efficient control systemdevelopment method based on real-time virtualhardware-in-the-loop simulation is presented. Insteadof developing and testing control system afterphysical prototype is produced, it can be conductedimmediately after concept of controlled system isdesigned by developing a pure software real-timesimulator. In this simulator, all the hardware isvirtual. The appearance and operation of these virtualhardware is programmed to imitate correspondingphysical prototypes so as to implement real-timevirtual hardware-in-the-loop simulation. So Thecontrol software and the control logic of controlsystem can be developed quickly. And the sourcecode is compatible with both the simulation platformand physical platform. Moreover, It’s much easier andfaster to debug and test the control logic in such away. An example of a large and high-groove densityruling engine is provided to explain detailimplementation of the proposed method.

FP4.4 P0604 1700 - 1720Design and Development of a 4kW SMPSPower Supply with Long Term Short CircuitCapabilityKosala JAYASUNDARA, Salinda TENNAKOONUniversity of Moratuwa

The design of high power AC-DC converters withhigher efficiency is a challenging task, specially withthe consideration of the long term short circuitwithstand capabilities. This kind of power supplies arecommonly used in Arc welding power supplies andother variable load supplies. This paper discussesabout the design and implementation of a 4kW AC-DC converter with output voltage stability, currentlimiting and the long term short circuit withstandcapability. Also a special type of snubber circuit isdiscussed. Index Terms — AC-DC converters,snubber design, short circuit protection, weldingtransformer

FP4.5 P0925 1720 - 1740A Unified Controller Design using two DifferentTypes of Optimal Reset Control Laws for HDDServo SystemsHui LI, Youyi WANGNanyang Technological University

This article proposes a unified reset control designmethod to achieve both track-seeking and track-following performance in hard disk drive servosystems. It involves a discrete Linear QuadraticRegulation (LQR) reset controller and an H2 resetcontroller. The LQR reset controller is much capableof improving transient response performance which

120

includes fast raise time and little overshoot. Thediscrete H2 reset controller performs a betterdisturbance rejection performance which is moresuitable for track following stage. The unified controlmakes a soft switch from LQR reset controller to H2reset controller, which avoids sharp vibrations.Simulation shows that the unified reset controllerinherits the merits of both discrete LQR resetcontroller and discrete H2 reset controller, andachieves both swift seeking and accurate followingperformance.

FP4.6 P0955 1740 - 1800Model Predictive Control Based EfficientOperation of Battery Energy Storage Systemfor Primary Frequency ControlMuhammad KHALID, Andrey V SAVKINThe University of New South Wales

This paper presents a method for an efficientoperation of a battery energy storage system (BESS)associated with frequency control problem. A controlsystem model is proposed to simulate the BESS forfrequency control application. A controller based onmodel predictive control (MPC) is designed for thereliable operation of the BESS for primary frequencyregulation. A frequency prediction model based onGrey theory is also designed to optimize theperformance of our predictive controller. The methodis tested using real measurements from a real powergrid in the presence of multiple and realistic physicalsystem constraints. The simulation results depict theeffectiveness of the proposed frequency regulationscheme.

Session FP5Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Pelican

Identification and Estimation IIChairs : Jia Yiing, Patricia WONG

Nanyang Technological UniversitySalim AHMEDQatar University

FP5.1 P0687 1600 - 1620Piecewise Linear Approximation of SinusoidalInputs for Process IdentificationSalim AHMEDQatar University

A new approach for simultaneous estimation ofparameters and the time delay of continuous timetransfer function models using piecewise linear inputexcitations is presented. A sinusoidal input can beapproximated as a piecewise linear signal.Approximation equation for sinusoids has beenprovided and estimation equation for first ordertransfer functions has been derived which can begeneralized for higher order models. Simulationresults are presented to demonstrate the efficacy ofthe method. Identification results of a simulatedjacketed reactor are presented to demonstrate theapplicability of the proposed algorithm.

FP5.2 P0554 1620 - 1640Identification and Control of NonlinearSystems using Piecewise Affine ModelsChow Yin LAI, Cheng XIANG, Tong Heng LEENational University of Singapore

Piecewise affine model is a useful tool forapproximating nonlinear systems. In this paper, wefirst propose a procedure for obtaining the piecewiseaffine ARX models of nonlinear systems. Twoparameters which fully characterize a piecewise affineARX model, namely the parameters of the locallylinear/affine subsystems, as well as the partitions ofthe regressor space, will be estimated, the formerthrough a least-squares based identification methodusing multiple models, and the latter using standardprocedures such as neural network classifier orsupport vector machine classifier. Based on thepiecewise affine ARX model of the nonlinear system,we then proceed to derive a model-based controllerto control the system for reference tracking.Simulation studies show that our algorithm canindeed provide accurate piecewise affineapproximation of nonlinear systems, and that theproposed controller provides good trackingperformance.

FP5.3 P0614 1640 - 1700Nonparametric Modeling of Glucose-Insulin

121

Process in IDDM Patient using Hammerstein-Wiener ModelArpita BHATTACHARJEE, Anindita SENGUPTA, AshokeSUTRADHARBengal Engineering & Science University, Shibpur

This paper deals with an identification problem ofmodeling a nonlinear dynamic system of multivariableglucose-insulin process in an IDDM patient. Out ofmany model structures that can represent a nonlinearprocess effectively; the Hammerstein-Wiener modelhas attracted a lot of attention. The present workproposes a generalized identification method ofHammerstein-Wiener model from the input-outputdata of multivariable nonlinear glucose-insulinprocess. The present algorithm consists of a three-block (LNL) realization. For the multivariable system,the first and third blocks are standard impulseresponse filter (IRF) realization applied to anequivalent linear system using adaptive recursiveleast square (ARLS) algorithms. In the second block,i.e. the nonlinear part, ARLS algorithms have beenused to solve up to second order kernels of Volterraequations with extended input vector consisting ofcross components as well. The input-output datataken from the simulated nonlinear process havebeen used to identify the system with a filter memorylength of M=3 and the validation results have showngood fit and in concordance with predicted output.

FP5.4 P0764 1700 - 1720Approximation by Discrete HermiteInterpolationFengmin CHEN, Jia Yiing, Patricia WONGNanyang Technological University

For a function defined on the discrete interval, wedevelop a class of quintic discrete Hermite interpolatethat involves only differences. Further, explicit errorbounds are offered in the form of the inequality,where the constants are explicitly provided. We alsoestablish the two-variable Hermite interpolate as wellas the related error analysis for a function defined ondiscrete intervals. Four numerical examples arepresented to illustrate the actual construction of thediscrete Hermite interpolates, the actual errors arealso computed to compare with the error boundsobtained.

FP5.5 P0765 1720 - 1740Approximation by Discrete Spline InterpolationFengmin CHEN, Jia Yiing, Patricia WONGNanyang Technological University

For a function defined on the discrete interval, wedevelop a class of quintic discrete spline interpolatethat involves only differences. Further, explicit errorbounds are offered in the form of the Inequality,where the constants are explicitly provided. Threenumerical examples are presented to illustrate theactual construction of the discrete spline interpolates,the actual errors are also computed to compare withthe error bounds obtained.

FP5.6 P0857 1740 - 1800Optimal Estimation of the Central Normal of aHeliostat using UKFFen XU, *Wenfeng LIANGNorth China University of Technology*Himin Solar Energy Group Co. Ltd

the aiming accuracy of the optical axis of aheliostat directly affects the working efficiency of aconcentrating solar tower plant. Due to the size ofheliostat, it is difficult to measure the orientation ofthe optical axis directly during the tracking process.With a computer-vision based measuring system, thecenter of reflecting beam of a heliostat can bedetected, but the result is disturbed by internal andexternal interferences. This paper presents a methodto estimate the orientation of optical axis of atracking heliostat using unscented kalman filter basedon the measured beam center data. Simulationresults and estimations based on actualmeasurements are presented in the paper.

122

Session FP6Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Oriole

Robot Sensing, Perception and DataFusionChairs : Vaidehi VIJAY

Anna University, ChennaiSen ZHANGSingapore Polytechnic

FP6.1 P0894 1600 - 1620Sound Source Target Localization System ofMobile RobotLinan ZU, Peng YANG, Hao SUN, Lingling CHENHebei University of Technology

Target localization system is one of the basicaspects for mobile robot navigation. One of the mostpopular navigation way is the navigation based onvisual. But, this way is limited by the light and thefield of view so that it can’t accurately detect thetargets in special environments. To resolve the aboveproblem of target localization way based on visual,this paper has studied the sound source targetlocalization system. The key technologies of thissystem have been discussed. They are the structureof the microphone array, the multi acoustic sourceseparation in dynamic environment, sound sourcelocalization method in mixed noise environment, thetime delay estimation method in the mixed-noiseenvironment, multi-sensor information fusion, and soon. Finally, an auditory system based on microphonearray was found to locate the sound source target of3-D space and some emulational experimentsvalidate the correctness of this auditory system andthe accuracy and real-time of the sound sourcelocalization.

FP6.2 P0992 1620 - 1640Binaural Speaker Recognition for HumanoidRobotsKarim YOUSSEF, Sylvain ARGENTIERI, Jean-LucZARADERUniversité Pierre et Marie Curie

In this paper, an original study of a binauralspeaker identification system is presented. The stateof the art shows that, contrarily to monaural andmulti-microphone approaches, binaural systems arenot so much studied in the specific task of automaticspeaker recognition. Indeed, these systems aremostly used for speech recognition, or speakerlocalization. This study will focus on the benefits ofthe binaural context in comparison with monauraltechniques. It demonstrates the interest of thebinaural systems typically used in humanoid robotics.The system is first tested with monaural signals, andthen with a binaural sensor, in many signal to noiseratios, speech durations and speaker directions. Up to

11 percent of improvement in recognition ratios of 23ms frames can be obtained. The used database is aset of audio tracks recorded for 10 speakers, andfiltered by HRTFs to obtain binaural signals in thedirections of interest, for the binaural training andtesting steps. This way, we study the sensitivity ofthe system to the speaker's location in anenvironment where a maximum of 10 speakers ispresent.

FP6.3 P1046 1640 - 1700A System for Reconstruction from Point Cloudsin 3D: Simplification and Mesh RepresentationLyuba ALBOUL, Georgios CHLIVEROSSheffield Hallam University

In this paper we present a complete system foracquisition of fused (textured) point clouds in 3D,from a Laser Range Finder (LRF) and a CCD camera.Furthermore, we describe an approach to build andprocess the resulting models, including their pre-processing and mesh simplification. This approachallows manipulating the resulting data structure intoconsistent geometric representations, which can befurther adapted based on user requirements. Theadvantage of our system is that of low computationalcost, ease of use and accuracy in the representationof the environment, even without prior datasmoothing.

FP6.4 P0958 1700 - 1720Multi-class Classification for Semantic Labelingof PlacesLei SHI, Sarath KODAGODA, Gamini DISSANAYAKEUniversity of Technology, Sydney

Human robot interaction is an emerging area ofresearch, where human understandable roboticrepresentations can play a major role. Knowledge ofsemantic labels of places can be used to effectivelycommunicate with people and to develop efficientnavigation solutions in complex environments. In thispaper, we propose a new approach that enables arobot to learn and classify observations in an indoorenvironment using a labeled semantic grid map,which is similar to an Occupancy Grid likerepresentation. Classification of the places based ondata collected by laser range finder (LRF) is achievedthrough a machine learning approach, whichimplements logistic regression as a multi-classclassifier. The classifier output is probabilisticallyfused using independent opinion pool strategy.Appealing experimental results are presented basedon a data set gathered in various indoor scenarios.

FP6.5 P1091 1720 - 1740Mobile Sensing and Simultaneously NodeLocalization in Wireless Sensor Networks forHuman Motion TrackingSen ZHANG, *Wendong XIAO, **Jun GONG,***Marcelo H ANG JR, ***Chen Khong THAM, RonnyQuin Fai THAM

123

Singapore Polytechnic*Institute for Infocomm Research**Northeastern University***National University of Singapore

This paper exploits optimal position of the mobilesensor to improve the target tracking performance ofwireless sensor networks and simultaneously localizeboth of the static sensor nodes and mobile sensornodes when tracking the human motion. In ourapproach, mobile sensors collaborate with staticsensors and move optimally to achieve the requireddetection performance. The accuracy of final trackingresult is then improved as the measurements ofmobile sensors have higher signal-to-noise ratiosafter the movement. Specifically, we cansimultaneously localize the mobile sensor and staticsensors position when localizing the human's positionbased on augmented extended Kalman filters (EKF).In the algorithm, we develop a sensor movementoptimization algorithm that achieves near-optimalsystem tracking performance. We also presented ansensor nodes management scheme in order todeduce the computation complexity when localizingthe static sensor nodes. The effectiveness of ourapproach is validated by extensive simulations.

FP6.6 P0472 1740 - 1800Study of Handoff Performance in a MobileWiMAX NetworkVaidehi VIJAY KUMAR, Poorani MAnna University, Chennai

Various attempts are being made to integrate thewidely deployed disparate terrestrial wirelessnetworks in order to provide global roaming. Amongthe attempts, the handoff is a major concern since itis the basis for any session continuity. This paperaims to provide a handoff in Mobile WiMAX network(IEEE 802.16e). Mobile WiMAX is a fast growingbroadband access technology that enables low-costmobile Internet applications, and realizes theconvergence of mobile and fixed broadband access insingle air interface and network architecture. Thispaper reveals some challenges that mobile users facewhen travelling across different base stations in aMobile WiMAX environment. One of the main goalswhen traversing between the networks, as a mobileuser, is to ensure the loss of data or loss ofconnectivity does not occur as a result of handoff.This paper aims to study the handoff latency andthroughput performance with respect to differentvelocities and focuses on simulation approach forconducting investigative studies for this framework.

Session FP7Date : Friday, 10 December 2010Time : 1600 - 1800Venue : Bluebird

Robot Control IIChairs : Danwei WANG

Nanyang Technological University

FP7.1 P0268 1600 - 1620Experimental Investigation of a Path FollowingController for Planar Snake RobotsPål LILJEBÄCK, *Idar U HAUGSTUEN, *Kristin YPETTERSENSINTEF ICT*Norwegian University of Science and Technology

This paper considers path following control ofsnake robots along straight paths. A controller isproposed which, under the assumption that theforward velocity of the snake robot is nonzero andpositive, guarantees K-exponential stability of thedistance between the snake robot and the desiredpath and also K-exponential stability of the headingof the robot with respect to the direction of the path.The performance of the path following controller isinvestigated through experiments with a physicalsnake robot. The experiments show that theproposed controller successfully steers the snakerobot towards and along the desired straight path.

FP7.2 P0515 1620 - 1640Centrifugal Force Compensation of a Two-Wheeled Balancing RobotMishari ALARFAJ, George KANTORCarnegie Mellon University

Dynamically balancing, two-wheeled robots withhigh centers of gravity have been researched overthe past few years in order to tackle the restrictionsof robots in human environments. These robots aredesigned to have the same dimensions as humansand be able to actively maintain stability. Howeverthese systems are only able to compensate in thefore-aft directions and are limited to slow motionswhen turning in order to prevent tipping due tocentrifugal forces. In this paper we present a two-wheeled balancing robot whose body is an actuatedfour-bar linkage that can be used to lean. This allowsit to control the position of its center of mass in thesideways direction. An overview of the robot conceptis presented, and controllers are designed to balance,turn, and lean. Results are presented in dynamicsimulation.

FP7.3 P0658 1640 - 1700Design of Length Measuring System forStewart Platform using New ForwardKinematics SolutionHamid REZVANI DASTGERDI, *Mehdi KESHMIRI

124

Sakht Tajhizate Sepahan- MAPNA Group*Isfahan University of Technology

In this paper, based on new presented approachfor forward kinematic problem, length measuringsystem for Stewart platform is designed. The analysisis performed by using a software platform whichmodels the system dynamics and control. A typicalfeedback linearization method is used to design thecontroller system and performance of designedcontroller is studied numerically with respect tomeasuring error and parametric uncertainties. Toprovide required parameters for controller, a newforward kinematic solution is designed and comparedwith Newton-Raphson method numerically. Based onend effector deviations from desired trajectory,actuator length measuring system is compared withmoving platform measuring system, which results inmore reliable trajectory tracking of controller in caseof actuator length measuring system.

FP7.4 P0662 1700 - 1720Robust Path Following Control of anUnmanned BoatZhenyu YUBeijing Jiaotong University

Robust path following capability is mandatory foran unmanned boat to perform tasks autonomously.This paper presents a mixed H∞/H2 based controlwhich enables the boat to follow paths withconnected straight line segments and circular arcs inthe absence of yaw angle measurement. Theperformance of the strategy is evaluated in field testand will be discussed in the paper.

FP7.5 P0724 1720 - 1740Dual-loop Adaptive Decoupling Control forSingle Wheeled Robot - Based on Neural PIDControllerXiaogang RUAN, Qiyuan WANG, Naigong YUBeijing University of Technology

This paper proposes a dual-loop adaptivedecoupling control method based on single neuronPID controller (DADC-SNPID), balancing the singlewheeled robot. A unique mechanical and controlhardware structure of single wheeled robot isdesigned and its simplified mathematical model isestablished using the Newton-Euler equationsaccording to the actual parameters of the designedrobot. A new balance and motion control method isproposed as dual-loop adaptive decoupling controlmethod based on single neuron PID controller tocontrol the robot. Results of simulation and physicalexperiments are conducted to illustrate effectivenessof the designed robot system and its controller undercertain conditions.

FP7.6 P0897 1740 - 1800Trajectory Generation of Straightened KneeWalking for Humanoid Robot iCubZhibin LI, Nikos G TSAGARAKIS, Darwin G

CALDWELL, *Bram VANDERBORGHTItalian Institute of Technology, Genova*Vrije Universiteit Brussel

Most humanoid robots walk with bent knees,which particularly requires high motor torques atknees and gives an unnatural looking. It is thereforeessential to design a control method that produces amotion which is more energy efficient and naturalcomparable to those performed by humans. In thispaper, we address this issue by modeling the virtualspring-damper based on the cart-table model. Thisstrategy utilizes the preview control, which generatesthe desired horizontal motion of the center of mass(COM), and the virtual spring-damper for generatingthe vertical COM motion. The theoretical feasibility ofthis hybrid strategy is demonstrated in Matlabsimulation of a multi-body bipedal model. Knee jointpatterns, ground reaction force (GRF) patterns, COMtrajectories are presented. The successful walkinggaits of the child humanoid "iCub" in the dynamicsimulator validate the proposed scheme. The jointtorques required by the proposed strategy arereduced, compared with the one required by the cart-table model.

125

Session FIDate : Friday, 10 December 2010Time : 0930 - 1800Venue : Foyer

Interactive Session III

FI.1 P0378Adaptive Generalized Minimum VarianceControl for TCP/AQMRoohollah BARZAMINI, Masoud SHAFIEEAmirkabir University of Technology, Tehran

A new controller AQM strategy for dynamicallyvarying TCP/AQM networks is proposed. Adaptivegeneralized minimum variance is Combination of online parameters estimations and generalizedminimum variance (GMV). The GMV is a suitablecontrol method especially in the presence ofmeasurement noise, but when the process dynamicschanges, the controller performance declines. Toovercome this problem, the adaptive GMV controlapproach is recommended. Controller performancewas evaluated in two modes, TCP network withunknown parameters and TCP network with timevarying parameters. Simulation results indicate thatthe adaptive GMV AQM (in either case) has been ableto keep queue length around the desired point.

FI.2 P0304Variable Structure Control with FeedforwordCompensator for Robot Manipulators subjectto Load UncertaintiesSuolin DUAN, Lanping CHEN, Zhenghua MA, GuirongLUJiangsu Polytechnic University

A robust control method based on equivalentcontrol of Variable Structure Control (VSC) withfeedforward compensator for robotic manipulatorssubject to load uncertainties is presented. Thefeedforward compensator is used to improve trackingprecision, and the feed-sequence compensator withthe Proportional Integral (PI ) regulator to thesliding-mode function is used for improving thedynamical performance of sliding mode state in VSCand attenuating the chattering phenomena inconventional sliding-mode state. The experimentalresults in GR-2 robot manipulators with two degreesof freedom verify the effectiveness of the presentedcontroller.

FI.3 P0385Robust Coordinated Control of Hot Strip MillMultivariable SystemKaixiang PENG, Jie DONGUniversity of Science and Technology Beijing

Hot strip mill is a typical process with highaccuracy and high speed. The effective approach toenhance the products competitiveness is to improve

the quality and quantity by means of advancedcontrol strategy. A multivariable system comprised ofgauge and mass flow of hot strip mill is analyzed inthe article. A state-space model about the gauge andmass flow for one stand and inter-stand in a hot stripmill is set up as well. New steady and coordinatedcontrol approach on the basis of LQG and H infiniteof hot strip mill is proposed and the robustcharacteristics such as perfect robust stabilization,strong disturbance attenuation and coordinationproperties are better in the latter. Simulation resultsshow the validity of analysis, design and control.

FI.4 P0775Nonlinear Sliding Mode Observers for FaultReconstruction and State EstimationsKalyana C VELUVOLU, *Yeng Chai SOHKyungpook National University*Nanyang Technological University

In this paper, we shall examine the design ofsliding mode observers for Lipschitz nonlinearsystems for state and faults/unknown inputestimations. The robust terms or the switching termsare designed such that the faults are tracked by theirrobust terms and so can be reconstructed from thesliding mode. The stability condition for the reducedorder system is analyzed and the feedback gain isdesigned such that the reduced order system isstable. An application example to robotic manipulatoris examined to demonstrate the effectiveness of theproposed method in reconstruction of unknowninputs/faults.

FI.5 P0986Sparse Structured H∞ Filter Design with forLinear SystemsWei Wei CHE, Yan-Ping LIShenyang University

This paper presents a study on the problem ofdesigning non-fragile H∞ filters with sparse structurefor linear discrete-time systems. The filters to bedesigned with sparse structure are assumed to bewith additive gain variations, which are resulted fromfilters implementations. Firstly, a class of sparsestructures based on the observable standard form isspecified from a given fully parameterized $H∞ filter.Then, an LMI-based procedure for designing non-fragile H∞ filters with the sparse structure is provided.The resulting design guarantees the augmentedsystem asymptotically stable and the H∞ attenuationlevel less than a prescribed level. A numericalexample is given to illustrate the proposed method.

FI.6 P0493

Robust H∞ Model Reduction for a Class ofNonlinear Time-varying Systems with TimeDelaysWeiwei ZHAO, Xiao HE, Yindong JITsinghua University

126

In this paper, we are concerned with the H∞model reduction problem for a class of nonlineartime-varying systems with parameter uncertaintiesand time delays in a discrete time framework. Byintroducing a proper Lyapunov function, we establisha sufficient condition ensuring the asymptoticallymean square stability of the associate error systemwith a prescribed H∞ error performance in the formof quasi-linear matrix inequality, based on which theparameters of the reduced-order model can also bedetermined. Finally, a numerical example is includedto show the effectiveness of the developedtechnique.

FI.7 P1088Multi-objective Robust PID Controller Tuningusing Multi-objective Differential EvolutionShizheng ZHAO, Bo-Yang QU, SuganthanPONNUTHURAI NAGARATNAM, *Iruthayarajan M.WILLJUICE, *Baskar SNanyang Technological University*Thiagarajar College of Engineering, Madurai

PID controller has been widely applied inengineering area. In this paper, multi-objectivedifferential evolution (MODE) is used to design amulti-objective robust PID controller for two MIMOsystems known as distillation column plant andlongitudinal control system of the supermaneuverable F18/HARV fighter aircraft. Multi-objective robust PID controller problem is formulatedby minimizing integral squared error (ISE) andbalanced robust performance criteria. Theperformance of the optimum PID controllers thatobtain by MODE is compared with performancereported in literature by other methods in terms ofthe sum of ISE and balanced robust performancecriteria. The results show that the PID controllersobtained by MODE can outperform various optimalPID controllers reported in literature.

FI.8 P1095Automatic Take-off Control System forHelicopter - An H∞ Apporach

Sundaram SURESH, *Pushpinder KASHYAB, *MasqueUn NABINanyang Technological University*Indian Institute of Technology, Delhi

The paper describes the development of anautomatic take-off system for helicopter. Design ofautomatic take-off control system is mandatory formanned as well as unmanned helicopter, whichreduce the burden of pilot and increases the flyingquality of the vehicle. In this paper, we present anH∞ based automatic take-off system for helicopter.For this purpose, we consider typical four-bladedhelicopter with conventional control. A single robustcontroller for altitude and attitude tracking aredesigned at single hovering altitude such thatcontroller stabilized the closed loop system at allhovering altitude. The performance of the controlleris evaluated in the presence of sensor noise and

moderate vertical gust.

FI.9 P0142Self-learning Continuous ControllersOtto CERMAN, Petr HUSEKCzech Technical University in Prague

In this paper different approach of adapta- tionmechanism for fuzzy model reference learning con-trol (FMRLC) method will be introduced. In contrastto original method the proposed procedureguarantees continuity of the fuzzy controller that inreward results in its smooth input-output behaviour(mapping) that eliminates undesirable abruptchanges in the control signal. Additionally, theproposed method shows larger robustness toresponses of initial conditions and vari- ous referencesignals. The advantages of the proposed modificationare presented on control of magnetic sus- pensionsystem. From result comparing rule bases afteradaptation there is absolutely smooth control surfacein case of proposed method FMRLC in comparisonwith original FMRLC. With proposed procedure it wasachieved similar result regulation to original FMRLCbut with more than 4x smaller adaptation gain.

FI.10 P0333A Comparative Analysis of LQR and Fuzzy LogicController for Active Suspension using Half CarModelFaried HASBULLAH, Waleed F FARISInternational Islamic University Malaysia

The application of the Linear Quadratic Regulator(LQR) and Fuzzy Logic Controller (FLC) in the field ofactive vibration isolation for a vehicle suspensionsystem is presented and the performances of the twoactive controllers are compared with the passivesystem. The performance of the LQR controller isbetter than the FLC at the expense of control force.The performances of both active controllers arebetter than the passive system in terms of settlingtime.

FI.11 P0488Multidimensional Minimization TrainingAlgorithms for Steam Boiler High TemperatureSuperheater Trip using Artificial IntelligenceMonitoring SystemFiras BASIM ISMAIL, Hussain H AL-KAYIEMUniversity Teknologi PETRONAS

Steam Boilers are important equipment in powerplants and the boiler trips may lead to the entireplant shutdown. To maintain performance in normaland safe operation conditions, detecting of thepossible boiler trips in critical time is crucial. As apotential solution to these problems, an artificialintelligent monitoring system specialized in boilerhigh temperature superheater trip has beendeveloped in the present paper. The BroydenFletcher Goldfarb Shanno Quasi-Newton (BFGS QuasiNewton) and Levenberg-Marquardt (LM) have been

127

adopted as training algorithms for the developedsystem. Real site data was captured from MNJ coal-fired thermal power plant-Malaysia. Among threepower units in the plant, the boiler high temperaturesuperheater of unit one was considered. Anintegrated plant data preparation framework forboiler high temperature superheater trip with relatedoperational variables, have been proposed for thetraining and validation of the developed system. Bothone-hidden-layer and two-hidden-layers networkarchitectures are explored using neural network withtrial and error approach. The obtained results wereanalyzed based on the Root Mean Square Error fordeveloped intelligent monitoring system.

FI.12 P0418Integration of INS and GPS using Radial BasisFunction Neural Networks for VehicularNavigationMalleswaran M, Angel Deborah S, Manjula S, *VaidehiVIJAY KUMARAnna University, Tirunelveli*Anna University, Chennai

Navigation systems used in recent days rely mainlyon Kalman filter to fuse data from global positioningsystem (GPS) and the inertial navigation system(INS). In common, INS/GPS data fusion providesreliable navigation solution by overcoming drawbackssuch as signal blockage for GPS and increase inposition errors with time for INS. Kalman filteringINS/GPS integration techniques used in present dayshave some inadequacies related to the stochasticerror models of inertial sensors, immunity to noise,and observability. This paper aims to introduce a newsystem integration approach for fusing data from INSand GPS utilizing artificial neural networks (ANN). Amulti-layer perceptron ANN has been recentlysuggested to fuse data from INS and differential GPS(DGPS). Though the integrated system using multi-layer perceptron scheme improves the positioningaccuracy, it has shortcomings like complexity withrespect to the architecture of multi-layer perceptronnetworks and limitation of online training algorithm toprovide real-time capabilities. This paper, therefore,proposes the use of an alternative ANN architecture.The proposed architecture is based on radial basisfunction (RBF) neural networks, which generally havesimpler architecture and faster training proceduresthan multi-layer perceptron networks. The RBF-ANNmodule is trained to predict the INS position errorand provide accurate positioning of the movingvehicle.

FI.13 P0498Environmental Monitoring and Air-conditioningAutomatic Control with Intelligent BuildingWireless Sensor NetworkTao ZHENG, Yajuan QIN, Deyun GAO, HongkeZHANGBeijing Jiaotong University

Wireless sensor network (WSN) is the connection

between the physical world and mankind.Particularly, environmental monitoring and devicesautomatic control of intelligent building based onwireless sensor network is considered as one of themost crucial applications. It can perceive many kindsof environmental parameters and feedback controlinformation to some devices to provide comfortableenvironment to people. However, it is difficult todeploy a WSN in the buildings because there usuallyare many wireless LAN devices used in the buildings,which bring serious frequency interferences. In thispaper, we conduct a real intelligent building wirelesssensor network (IBWSN) for environmentalmonitoring and air-conditioning automatic control. Inorder to ensure the effectiveness of this system,actual spectrum analysis is developed. Performanceevaluation proves that the presented IBWSN cansatisfy the needs of the proposed applications.

FI.14 P1067Movement Persuit Control of an OffshoreAutomated Platform via a RAM-based NeuralNetworkHoracio L FRANÇA, João Carlos SILVA, *Massimo DEGREGORIO, **Omar LENGERKE, Max S DUTRA,Felipe M G FRANÇAUniversidade Federal do Rio de Janeiro*Istituto di Cibernetica - CNR**Universidad Autónoma de Bucaramanga

The reproduction of the movements of a ship byautomated platforms, without the use of sensorsproviding exact data related to the numeric variablesinvolved, is a non-trivial matter. The creation of anartificial vision system that can follow the cadence ofsaid ship, in six axes of freedom, is the goal of thisresearch. Considering that a real time response is arequisite in this case, it was decided to adopt aBoolean artificial neural network system that couldidentify and follow arbitrary interest points that coulddefine, as a group, a model of the movement of anobserved vessel. This paper describes thedevelopment of a prototype based on the Booleanperceptron model WiSARD (Wilkie, Stonham andAleksander's Recognition Device), that is beingimplemented in the C programming language on adesktop computer using a regular webcam as input.

FI.15 P0125Opportunistic 3D Trajectory Generation for theJPL Aerobot with Nonlinear TrajectoryGeneration MethodologyWeizhong ZHANG, Tamer INANC, *Alberto ELFESUniversity of Louisville*California Institute of Technology

NASA is supposed to implement a sustainable andaffordable human and robotic program to explore thesolar system and beyond as it is the first goal of ThePresident’s Vision for U.S. Space Exploration. Therobotic exploration across the solar system consists ofexploring Jupiter’s moons, asteroids and other bodiesto search for evidence of life, and to understand the

128

history of the solar system. Trajectory generation fora robotic vehicle is an essential part of the totalmission planning. To save energy by exploitingpossible situation such as wind will assist a roboticexplorer extend its life span and perform tasks morereliably. In this paper, we propose to utilize NonlinearTrajectory Generation (NTG) methodology togenerate 3D opportunistic trajectories for an Aerobotby exploiting wind. The Aerobot is dynamicallycontrolled by three propellers which are respectivelyparallel to the local three Cartesian axes. Constraintsfor the Aerobot control are derived from Euler-Lagrange equations since the Aerobot satisfies withthe Lagrange-D'Alembert principle. The newproposed Aerobot model takes the aerodynamics intoaccount. The results show that NTG can take theadvantage of wind profiles to save significant energyfor the defined goal.

FI.16 P0940Surface-tracking of a 5-DOF ManipulatorEquipped with Tactile SensorsJingguo WANG, Yangmin LIUniversity of Macau

Tracking a surface via the end-effector of amanipulator is a tough issue, one not only shouldimplement the position and force control , but alsomonitor and sense the actual contacting statebetween the end-effector and the objects surface. Inthis paper, we present an approach integrating thetactile sensing with force-torque information as thefeedback to control the manipulator tracking asurface. Hybrid impedance control method is appliedto follow both the position and force trajectories. Atthe same time, the posture of the end-effector isexpected to keep horizontal. With the feedback oftactile sensing data such as contact state, contactarea and so on, several strategies of tactile sensingfeedback are proposed to be included into the controlalgorithm. Simulations and two groups of realexperiments are made using a 5-DOF manipulatorequipped with force/torque and tactile sensors tocontact with hard and soft board respectively, andthe results are compared and analyzed. Theeffectiveness of the proposed strategies is validated.

FI.17 P1158Recurrent Neural Network Based TrackingControlZhao XU, Qing SONG, Danwei WANGNanyang Technological University

In this paper, a recurrent neural network (RNN)based robust tracking controller is designed for aclass of multiple-input-multiple-output (MIMO)discrete time nonlinear systems. The RNN is used inthe closed-loop system to estimate online unknownnonlinear system function. A multivariable robustadaptive gradient-descent training algorithm isdeveloped to train RNN. The proposed neural controlsystem guarantees the stability of the closed-loopsystem and good tracking performance is achieved.

FI.18 P0324Distributed Control of Mobile Robotic SensorNetworks for Multi-level Barrier CoverageTeddy M CHENG, Andrey V SAVKINThe University of New South Wales

We study a problem of K-barrier coverage byemploying a network of self-deployed, autonomousmobile robotic sensors.A distributed motioncoordination algorithm is proposed for the mobilerobotic sensors to address the coverage problem. Thealgorithm is theoretically developed based on somesimple consensus algorithms that only rely on localinformation. By applying the algorithm to the sensors,K layers of sensor barriers are formed between twogiven points. To illustrate the proposed algorithm,numerical simulations are carried out for a number ofscenarios.

FI.19 P0890A Distributed Multi-robot Adaptive SamplingScheme for Complex Field EstimationMuhammad F MYSOREWALA, *Dan O POPA, LahouariCHEDED, Mirza Salman BAIGKing Fahd University of Petroleum and Minerals,Dhahran*The University of Texas at Arlington

Monitoring widespread environmental fields is acomplex task that is of great use in many areas, suchas building models of natural phenomenon: e.g.moisture in a crop field, oil reservoirs, etc. Asuccessful monitoring of such spatio-temporallydistributed fields hinges upon the use of wirelesssensor networks which, through their distributednature, allow for an effective adaptive samplingprocedure to gather the statistical informationnecessary for field density estimation. The adaptivenature of the sampling procedure used embodies astrategy which selects the next sampling locationbased on the gathered statistical information, andwhich evolves with past measurements. This paperpresents a novel distributed multi-robot “Adaptivesampling algorithm”, which is an extension of thealgorithm proposed earlier for complex fieldestimation using a single-robot only. Newformulations of sensor fusion in a centralized,decentralized, federated-decentralized, anddistributed sensor network are presented for fielddensity estimation, and not just cloud boundarydetermination. A comparison of the variouscomputational loads involved is included. Simulationresults show that adding an efficient partitioning ofthe sampling area and parallel multi-robot samplingimproves the field reconstruction time. With N robots,more than an N-fold reduction in the number ofsampling times is observed. The federated anddistributed scheme also leads to an improvedcommunication and computational efficiency.

129

FI.20 P1085Interacting Multiple Models BasedClassification of Moving ObjectsJulien BURLET, Olivier AYCARD, Qadeer BAIGUniversity of Grenoble 1

In this paper, we present an approach performingobject behavior classification embedded in a complexand efficient perception method. This method,applied in dynamic outdoor environments using amoving vehicle equipped with a laser scanner, iscomposed of a local simultaneous localization andmapping (SLAM) with detection and tracking ofmoving objects (DATMO). While the SLAM isperformed by an implementation of incremental scanmatching method, the tracking if performed by aMultiple Hypothesis Tracker (MHT) coupled with anadaptive Interacting Multiple Models Filter (IMM). Theclassification process takes place in the filtering stageand is based on one of the key parameters of theIMM filter which is the Transition Probability Matrix(TPM) modeling objects motion transitions. It permitsto automatically classify object behavior and to reusethe classification output to enhance the predictionstep in the filtering process. The experimental resultson datasets collected from a Daimler Mercedesdemonstrator in the framework of the EuropeanProject PReVENT-ProFusion2 demonstrate thecapacity of the proposed algorithm.

FI.21 P0961Magician Simulator - A Realistic Simulator forHeterogeneous Teams of Autonomous RobotsAdham ATYABI, Tom A F ANDERSON, KennethTREHARNE, David M W POWERSFlinders University

We report on the development of a new simulationenvironment for use in Multi-Robot Learning, SwarmRobotics, Robot Teaming, Human Factors andOperator Training. The simulator provides a realisticenvironment for examining methods for localizationand navigation, sensor analysis, object identificationand tracking, as well as strategy development,interface refinement and operator training (based onvarious degrees of heterogeneity, robot teaming, andconnectivity). The simulation additionally incorporatesreal-time human-robot interaction and allows hybridoperation with a mix of simulated and real robots andsensor inputs.

FI.22 P0256An Ensemble Vector Median Filter for ColorImage DenoisingLing ZHONG, Yun ZHANG, Yan XINGGuangdong University of Technology

Focusing on noise reduction in color images, anensemble-based vector median filter (EVMF) isproposed, which integrates the technique of vectormedian filtering into the framework of ensemblelearning. The algorithm consists of two main steps:1) Base filtering step, where a contaminated image isfiltered by a group of vector median filters (i.e. base

filters) in parallel; 2) Meta-filtering step, where theoutputs of all the base filters are fed into a vectormedian filter (i.e. meta-filter) to obtain the finalresult. The effectiveness performance of EVMF isevaluated and compared with that of several non-ensemble vector median filtering algorithms throughthe simulation experiment. The experimental resultsdemonstrate that EVMF outperforms the non-ensemble filtering approaches on both noisesuppression and detail preservation.

FI.23 P1055Players Tracking and Ball Detection for anAutomatic Tennis Video AnnotationKosit TEACHABARIKITI, *Thanarat HCHALIDABHONGSE, Arit THAMMANOKing Mongkut's Institute of Technology Ladkrabang*Chulalongkorn University

This paper describes our algorithms for playerstracking and ball detection for an automaticbroadcast tennis video annotation. The systemdetects and tracks the players using a robust non-parametric procedure for estimating density gradientscalled the mean shift algorithm. The basic mean shifttracking algorithm assumes that the target object hasto separate sufficiently from background, but thisassumption is not always true especially whentracking is carried out in dynamic backgrounds suchas in sport videos. To cope with this problem, in ourproposed system, we embrace the motionsegmentation and use the 8x8x8 color histogram tobe feature distribution for mean shift tennis playerstracking. In order to determine the players’ actionsprecisely, the system also detect and track ballpositions using frame differencing as well as applyingsome correlation techniques to eliminate falsedetections. Based on both players’ motion patternsand ball positions, the system can precisely classifythe players’ action into backhand ground stroke andforehand ground stroke. Videos of broadcast tennisgames downloaded from the Internet have beentested. The results show our system is able toprecisely classify the player’s actions with 83.7%precision and 82% recall rates.

FI.24 P1070HOG Based Multi-Stage Object Detection andPose Recognition for Service RobotLi DONG, Xinguo YU, Liyuan LI, Jerry Kah Eng HOEInstitute for Infocomm Research

This paper develops a HOG-based multi-stageapproach for object detection and object poserecognition for service robots. This approach makesuse of the merits of both multi-class and bi-classHOG-based detectors to form a three-stage algorithmat low computing cost. In the first stage, the multi-class classifier with coarse features is employed toestimate the orientation of a potential target object inthe image; in the second stage, a bi-class detectorcorresponding to the detected orientation withintermediate level features is used to filter out most

130

of false positives; and in the third stage, a bi-classdetector corresponding to the detected orientationusing fine features is used to achieve accuratedetection with low rate of false positives. The trainingof multi-class and bi-class SVMs with their respectivefeatures in different levels is described. Experimentsin real-world environments have shown that theproposed method is much more accurate than thedetection method as it uses only multi-class detector.The proposed method is also much more efficientthan the detection method as it uses a bi-classdetector for each possible orientation. The approachworks well on the scenarios where the SIFT-baseddetector may fail. The method can achieve real-timeobject detection, localization, and pose recognition ona P4 2.4GHz PC.

FI.25 P1104Full Spectrum Vibration Suppression for VideoStabilizationWen-Jong LIN, Chuen Leong NG, Hong LUO,ZhingMing GONG, Guilin YANGSingapore Institute of Manufacturing Technology

Stable and clear video imaging system is crucial invarious industry applications for monitoring andremote control purposes. Severe vibrations across fullspectrum frequencies are inevitable in manyindustrial operations resulting in shaky and unclearimages. It may reduce the response time for operatorand also affect the throughput during the productionor monitoring process. On the other hand,commercially available cameras with built-in imagestabilization functions may not be sufficient toovercome this problem since they only deal withcamera shaking at small magnitude. To stabilizeimages captured under sever and full spectrumvibration condition, a vibration suppression systemwas developed. The system consists of a passivevibration mechanism and a digital image stabilizationtechnique. The eddy current damper inside thepassive vibration mechanism reduces high frequencyvibration, while the digital image stabilization shiftsthe image for low frequency vibration elimination.Trial results have demonstrated the effectiveness ofthe system’s capability to cover full vibrationspectrum. However, it is also found that it needs toincrease the degree-of-freedom (DOF) for thismechanism in the future instead of only single DOF.

FI.26 P0482Eye-Gaze Detection with a Single WebCAMbased on Geometry Features ExtractionNguyen Huu CUONG, *Huynh THAI HOANGCantho University*Ho Chi Minh City University of Technology

In this paper, we propose an efficient approach forreal-time eye-gaze detection from images acquiredfrom a web camera. The measured data is sufficientto describe the eye movement, because the webcamera is stationary with respect to the head. First,the image is binarized with a dynamic threshold.

Then geometry features of the eye image areextracted from binary image. Next using estimationmethod based on geometry structure of eye, wedetect the positions of two eye corners. After that,the center of iris is detected by matching between aniris boundary model and image contours. Finally,using the relative position information between thecenter of iris and the eye corners, base on therelationship between image coordinate and monitorcoordinate, the position where the eye is looking atthe monitor is calculated. This system requires only alow cost web camera and a personal computer.Experimental results show that the proposed systemcan detect accurately eye movements in real-time.

FI.27 P0601Visual Interpretation of Natural PointingGestures in 3D Space for Human-RobotInteractionZhi LI, Ray JARVISMonash University, Melbourne

Visual interpretation of natural pointing gestures isessential in a human robot interaction scenario. Bothhands and head are involved in pointing behaviors.Given the color images acquired by a web cameraand the depth data by a TOF range camera, weperform visual tracking of the head and hands in 3Dspace. We investigate both the Head-Finger Line(HFL) and the forearm orientation as the estimationof the pointing direction. HFL is determined by 3Dpositions of the face and finger tip. Forearm directionis calculated using the PCA method in the RANSACframework. Their performances are evaluated andcompared in the experiments. Face direction and eyegaze orientation provide important cues regardingwhere the person’s attention is during a pointingoperation, which is proven helpful for eliminatingsome false estimations in our experiment.

FI.28 P0648Liver Cancer Identification based on PSO-SVMModelHuiyan JIANG, Fengzhen TANG, Xiyue ZHANGNortheastern University

This paper proposes a novel liver canceridentification method based on PSO-SVM. First, theregion of interest (ROI) is determined by Lazy-Snapping, and various texture features are extractedfrom ROI. Afterwards, F-score algorithm is applied toselect relevant features, based on which liver cancerclassifier is designed by combining parallel SupportVector Machine (SVM) with Particle SwarmOptimization (PSO) algorithm. PSO is used toautomatically choose parameters for SVM, and theadvantage is that it makes the choice of parametermore objective and avoids the randomicity andsubjectivity in the traditional SVM whose parametersare decided through trial and error. The experimentresults on real-world datasets show that theproposed parallel PSO-SVM training algorithmimproves the prediction accuracy of liver cancer.

131

FI.29 P0114Novel Web Cache ArchitectureMeenakshi KALIA, Ridhi SOODShobhit University, Meerut

Web cache is system that is placed between theservers and the clients so as to retrieve the webobjects to the clients as quickly as possible. But thealgorithms present today are not so efficient to retainthe important objects in the web cache. In this paperwe analyze several replacement algorithms used inweb cache like LFU, LRU and find that sometimesimportant links are also replaced. Whenever thoselinks are requested again, than that request has to bedirected to the server which results in wastage oftime. These findings guide us to propose a newunderlying architecture over which any of the desiredreplacement algorithms can be used. Besidereplacement factors like recency or frequency used inthe replacement we use other factors like Link Rankand Penetration Factor. We divided the web cacheinto five regions: Mature Region, Young Space,Micro-Cache, Non Popular Object Region and TrashBin.

FI.30 P0882Verification of Bank Cheque Images usingHamming MeasuresSankari M, Benazir M, *Bremananth RAMACHANDRANNerhu Institute of Engineering and Technology*Nanyang Technological University

The main objective of this paper is to verify bankcheques by using account number and accountholder’s signature present on the cheque image.Mainproblem is to exact localization of active regionsamong non-active contours in the image. Here, welocate the regions based on the prior knowledge ofCartesian coordinate space. It further involvesvarious steps such as Gray-Scale conversion,Segmenting contour, Inner-localization of accountnumber, Feature Extraction and Verification. Gray-scale conversion has been performed on chequeimages for reducing dimensionality of cheque size,which are required for further processing.Segmentation and localization are employed togetherto extract active-regions-of-interests such as accountnumber and signature. Furthermore, segmentedaccount numbers are obtained into individual digitsusing inner-localization. Feature Extraction isimplemented on both account number and signaturewith trained images using Hamming distancemeasures (HDM). Finally the verification is processedto identify the matching. Trained and testing featureswere compared.

FI.31 P1102A Novel Approach to Remove Redundant GaborWavelets for Family ClassificationMohammad GHAHRAMANI, Ngoc Minh DANG, EamKhwang TEOH, *Wei Yun YAU

Nanyang Technological University*Institute for Infocomm Research

Gabor Wavelets are widely used to extract facialfeatures since they are robust against illuminationand pose changes. Due to the limitation incomputational power, the common practice is todown-sample the face image to reduce number ofGabor features generated. As not all of the generatedGabor features are necessary, the main objective ofthis paper is to develop an efficient removal schemeof redundant filters in order to employ images withlarge dimension in face data processing. In particular,Genetic Algorithm is used to provide a computationaland fast selection of feature ensemble. The baseclassifiers are trained by the AdaBoost algorithm withthe Gabor feature set extracted from each singleGabor filter. By employing the joint diversity, GeneticAlgorithm is then applied to select the mostdiscriminate ensemble of classifiers followed by theoptimum decision making rule on the classifiersoutputs. The proposed approach is implemented inthe family classification problem which has largeintra-group variation. This method also allows us toselect more discriminate filters from higher scales andfiner orientations for those families with very youngchildren to improve the performance with the samecomplexity and calculation load of the conventionalGabor wavelet set.

132

REVIEWERS

AABBAS Vali UddinABBASI Abdul RehmanABDELHALIM BoutarfaACHOUR NouaraAHMAD ZainalAHMED ChemoriAHMED SalimAKBARIMAJD AdelAKMELIAWATI RiniALDEEN MohammadALI A B M ShawkatALVANITOPOULOS PetrosAMANATIADIS AggelosAMIRA AbbesAN JiyaoANDREW TeohAREEKUL VutipongARZANPOUR SiamakAZEVEDO Filipe

BBACK JuhoonBAI LinBAI XiaomingBAIG ZubairBAO GuanjunBARBOSA RamiroBENSAALI FaycalBERTUCCELLI LucaBHATTI AsimBI ShuhuiBIGDELI AbbasBIGLARBEGIAN MohammadBLUM JacquesBONNIFAIT PhillipeBOUSSARD ClémentBRANCO Costa

CCADONI MarinellaCAI ChengCAI ChenxiaoCAI WenchuanCAI YunzeCAMPS-VALLS GustavoCANTONI MichaelCAO CaopeiCAO KecaiCAO XiCAO XirenCAO YongcanCARVALHO JoseCHAI DouglasCHAKARAVARTHY VenkateswaranCHAN JianCHAN Y H, ChrisCHANDRASEKARAN VenkatachalamCHANG Dong EuiCHANG HongCHANG Hyeong Soo

CHANG Jeang LinCHANG Wen-ChungCHANGZHU ZhangCHAO HaiyangCHAUDHARY Dilip DCHE WeiweiCHECCHIN PaulCHEKIMA AliCHEN Chun-HaoCHEN HongCHEN JimingCHEN LiangCHEN MichaelCHEN ShaokangCHEN Tai-PangCHEN Xue-BoCHEN ZhiyongCHENG HaoCHENG ShanCHENG YongshengCHESI GrazianoCHI ZheruCHIN Sim YeeCHOI Yoon-HyukCHOUDHURY ShoukatCHOW W S, TommyCHU HenryCHUI C KCIUFUDEAN CalinCONTE GianpaoloCOSTA Antonio

DDA ShiDAI Shi-LuDAI YuchaoDAKSHINAMURTHI SivakumarDANIEL Paul JosephDARDEL MortezaDARGHAM JamalDAVOUDPOUR MaryamDE MADRID Angel PDECRAENE JamesDENG JeremiahDESOUZA GuilhermeDEY RajeebDIAS CustodioDING DouzhangDING KaiDING WeiDOKI KaeDONG JunfengDONG LijuDONG ZhaoyangDOS SANTOS MarcusDU RuoDU WenliDU YingziDUAN Hai-BinDUAN LiduanDUNNIGAN Matthew WDZIEKAN Lukasz

EEGHTESAD MohammadESMAEILI MOTLAGH Omid Reza

FFALLAHI MFAN YongFAN ZhenpingFANG FangFANG YongchunFARIS WaleedFENG ChangshuiFENG JuneFENG YingFERNÁNDEZ Antonio GiménezFERREIRA JoseFERREIRA MargaridaFERWORN AlexFIGUEIREDO LinoFOURQUET Jean-YvesFRANCO AnnalisaFRAUNDORFER FriedrichFREITAS GustavoFROM Pål JohanFU ChengFU HongFU Lixin

GG SainarayananGALEAZZI RobertoGALHANO AlexandraGAO DaoxiangGAO DeyunGAO GaoyongGAO XinwenGAO YonggaoGARCIA Gabriel JGARCIA-ROSA Paula BastosGARRIDO Aitor JGARRIDO IzaskunGE ShuzhiGE ZhiqiangGEORGOUDAS IoakeimGEORGOULAS ChristosGHOSE DebashishGIGUÈRE GyslainGOH Kah-OngGOH Wooi BoonGONÇALVES ManuelGOPALUNI BhushanGORDILLO FranciscoGRAVDAHL Jan TommyGU WenfeiGU XingshengGUAN YunqingGUNNERUD VidarGUO JiaxingGUO Zhongwei

133

GUO Zhongwen

HHALAS MiroslavHALIM IskanderHAN DongfangHAN JiandaHAN JinluHAN TaoHAN ZhengzhiHASAN GHASEMI MohammadHASSINI AbdelatifHAUGSTUEN Idar UHAUTÈRE NicolasHE JiazhouHE LianghuaHE ShupingHEIN AndreasHERATH Damith CHOLDEN ChristianHONG RichangHONG YiguangHORAN BenHOSSNY MohammedHOU JinHOU Saing PaulHOU TingHOU ZujunHOW Jonathan PHU ChaofangHU CunhuHU HucunHU SongyuHU XiangdongHU XiaomingHUA JianningHUANG Bin-JuineHUANG DanHUANG DongHUANG HaiboHUANG LeiHUANG RuiHUANG WeiminHUANG YanyanHUANG YulinHUDHA KhisbullahHUIMMIN LiHUSEK Petr

IINOUE TakahiroINSAURRALDE CarlosIOANNIDIS KonstantinosISMAIL Mohamed MIU Herbert H C

JJAKUBIAK JanuszJANCZAK AndrzejJANG Dae-SungJARA Carlos AJESUS IsabelJIA Wenjing

JIAN LiJIANG HaiboJIANG HuiyanJIANG JunJIANG LiJIANG WeiJIANG YuJIMENEZ Juan FranciscoJOHNSON-ROBERSON MatthewJONES Colin NeilJOSHI Maulin MJUNG Chihyun

KK R VenugopalKADIRKAMANATHAN VisakanKAEWKHAM-AI BoonsriKALIA MeenakshiKALYAN BharatKAMATA Sei-IchiroKANAVALLI AnitaKANKANHALLI MohanKANNO TaroKARIWALA VinayKARSITI Mohd NohKARTHIK NandakumarKAZMIERCZAK EdKHAN Masood MehmoodKHOO Sui YangKHORSHEED Mohammad SKIM Dong-HanKIM JaihieKODAGODA SarathKOMATSU NaohisaKONDO ToshiakiKONG AKONG LingheKONG ShulanKONG Xiang BingKONGPRAWECHNON WareeKONSTANTINIDIS KonstantinosKRAMER AnnikaKÜHNAPFEL ThorstenKÜHNLENZ KoljaKUMAR AjayKUMAR RajeshKUMAR Ratnesh

LLACROIX SimonLAGORIO AndreaLAI Chow YinLAMBERT AlainLAMBERT-TORRES GermanoLAN Guo-ChengLAN WeiyaoLAN YuanLANGLOIS DominicLAW N F, BonnieLE Hung LanLEE ChulheeLEE GordonLEE Hui MienLEE Jacky

LEE Sang-IlLEE SangjinLEE Suk-HanLEE Yeong-ChyiLEI FeiLEOPUTRA WilsonLI DanyongLI DaopingLI DequanLI DeweiLI FanLI JunLI LinLI LiyuanLI PengLI QimingLI QinLI ShuaiLI TaoLI XiangLI XiangpengLI Xiao HuaLI XiaoanLI XiaoguangLI XiaomingLI XuLI Yong-FuLI YongminLI ZhenLIANG TingtingLIAO Tai-ShanLILJEBÄCK PålLIM Kart LeongLIM Wei QuanLIMA MiguelLIN Chun-WeiLIN JipengLIN WeiLIN Wei-ChengLIN Xin-HuaLIU ChongLIU FengLIU FuchunLIU GangLIU HuiyangLIU JiLIU Jiang, JimmyLIU KexiuLIU LiuyanLIU ShuaiLIU ShuangLIU SuqingLIU SuyiLIU Wan QuanLIU WeihuaLIU XiaoxianLIU XiaoyanLIU Yi-HungLIU YijianLIU YuLOPES AntonioLOU YunjiangLOW Yoke Hean, MalcolmLU CongLU HongLU Jiwen

134

LU LiangLU Tien-FuLU XinjiangLU YaoLUN DanielLUO HongLUO MingLV Zhanggang

MMA BinMA HongjiMA Li-ShanMA YanzhuoMAHINDRAKAR Arun DMAIRE FredericMALAKHOV Oleg SMALCOLM Andrew AlexanderMAN ZhihongMANDAL BappadityaMANVI Sunilkumar SMANZANERA AntoineMANZIE ChrisMARCAIDA SilviaMARQUES ViriatoMARTIN PhilippeMARTÍNEZ-RAMÓN ManelMATSUO KazuyaMBAITIGA ZacharieMEI JieMEI PingMEI XiaoyanMENDEZ Gerardo MMENG XiangyuMETTIN UweMIKHAILOV AlexeiMISHRA PrabhakarMOHAMED Haider A FMONCRIEFF SimonMORATUWAGE DilukaMORENO-VALENZUELA JavierMORISSETTE LaurenceMOTA AlexandreMOURA OLIVEIRA PauloMUDUGAMNWA DamithMUHAMAD AMIN Anang HudayaMUÑOZ-MARÍ JMURAKAMI KoujiMUSCATO GiovanniMUSOROMY Zuwena

NNABI MNAIR GirishNANNI LorisNASSIRHARAND AmirNASUTION BennyNG Say TyamNG Teck ChewNGAMWIWIT JongkolNGUYEN Vu AnhNIAMSUP PiyapongNIEN H HNISHIO Kimihiro

NOMM SvenNQUYEN-DUC HiepNURGES Ulo

OOLIVEIRA TiagoONG Ee PingONG Sim HengORTIZ JesusOU XiaolingOZAWA RyutaOZBAY Hitay

PPACHECO LluísPALANIAPPAN KanthabhabhaPAN ShuwenPAN Tian-HongPANDIAN ShunmughamPANG Chee KhiangPARAMESWARAN LathaPARK Kang RyoungPATAN KrzysztofPATAN MaciejPATERNOSTRO MauroPENG DanPENG EnPENG KaixiangPENG ZhongboPEREIRA CarlosPÉREZ José AntonioPETLENKOV EduardPEURSUM PatrickPEYMANI EhsanPHAM Manh TungPHAM Trung DungPHUNG DinhPILLAI HarishPINHO MariaPINTO CarlaPIRES EduardoPISSANIDIS GeorgiosPOMARES JorgePOMMIER-BUDINGER ValériePONDA SameeraPU Xing-Cheng

QQI MingruiQI TianQIAN DianweiQIN ZengchangQIU XuyunQUAH Chee Kwang

RRAHMAN M K MRAJA AHMAD Raja KamilRAJPUROHIT Vijay SRAO AkshayRAO JinjunRATTANI Ajita

REIS CeciliaREN JianghongREN WeiREN YanREYES Napoleon HRICHARDSON TomROMERO MiguelROSARIO MaruricioRUBRICO Jose Ildefonso URÜTHER Matthias

SSAEED KhalidSAGARA ShinichiSAKAGAMI NSANG QianSANTOS PauloSATAPATHY SureshSATISH TadiparthiSAUER JürgenSCHMID RobertSCHMIDT KlausSEGHOUANE Abd-KrimSEKIMOTO MasahiroSENARATNE NamalSEQUEIRA JoãoSHANG FangSHARMA RahulSHE QingshanSHENG LiSHENG ZhiqingSHENOY ArunaSHI HongboSHIBATA MizuhoSHIH Ta-MingSILVA FilipeSILVA ManuelSIM TerenceSINGH Arun KumarSINGH Navdeep MSINGH ViijanderSIVARAMAKRISHNAN JanardhananSO H CSONG QiSONG Sang-HwaSONG XinminSONG YuSONG ZheSRINIVSAN BalaSRIVASTAVA SmritiSU HoushengSU Weidong, StevenSUBBARAMAN ShailaSUBODH BhandariSUN ChangshengSUN FumingSUN GangSUN Guo BingSUN HuiyingSUN JingSUN JinshengSUN MinhuiSUN ShaochaoSUN Yeow ChengSUN Ying

135

SUN ZhendongSUN ZhenguoSUN ZongyaoSURESH SundaramSUTRADHAR AshokeSVININ MikhailSWAIN Akshya

TTAGADE PiyushTAHARA KenjiTAI ZhixinTAKAGI KentaroTAKEMURA FumiakiTAMMI KariTAN Hock SoonTAN Su Lim, ForestTAN TeleTAN YouhuaTAN ZhiyuanTANG Chih-WeiTANG ZhenshanTANGUY NoelTAO JiTAO YifeiTAY Leng Phuan, AlexTEE ConnieTEOH Beng Jin, AndrewTERASHIMA KazuhikoTHAMBIPILLAI SrikanthanTIRDAD KayvanTIWARI ArunaTOMBARI FedericoTRAN Quang VinhTRICAUD ChristopheTSAKALIS Kostas

UUEMURA Mitsunori

VVASSEUR PascalVELUVOLU Kalyana CVERHAEGEN MichelVIJAY KUMAR VaidehiVIJAYANANDH RVILATHGAMUWA Mahinda

WWADA TakahiroWAN ChanglinWANG BinWANG CanWANG ChangjunWANG Chi-KaiWANG DanWANG DangxiaoWANG HaoWANG Hee LinWANG HongliangWANG HuaweiWANG Jiandong

WANG Jian-GangWANG KeWANG LiangshunWANG LichaoWANG LipoWANG LishengWANG LizhenWANG RuiWANG ShengWANG Shyi-WenWANG WWANG WeiWANG WeiqunWANG XiaofanWANG XiaolinWANG XinWANG Y LWANG YanWANG YouyiWANG YuanquanWANG YueWANG ZhengWANG ZhiyongWATAGODAKUMBURA ChandanaWEI BaoguoWELSH JamesWENG LiguoWEST GeoffWEYER ErikWIERING Marco AWIJESOMA Wijerupage SardhaWILLIAMS EdwardWONG Hau SanWONG ManWONGWIRAT OlarnWOON Wei-LeeWOUNGANG IsaacWU BuzhouWU Chih-HungWU JianWU JoannaWU JunWU JunfengWU Min-ThaiWU PanlongWU QiangWU QuanjunWU ShunxiangWU TaoWU YanWU YanhuaWU Zhongcheng

XXI BinXIA LinlinXIAO GaoxiXIAO JingXIAO JunXIAO LingfeiXIAO NanXIAO QinghanXIE XiangshengXIE XiaozhuXIE Xudong

XIE YuXIN XinXING ShusongXIONG GuangmingXU BinXU ChenfengXU DeXU JiaheXU JunXU MinXU YaojinXUE Xiangyang

YYAN FeiYAN JunYAN MaodeYAN ZhiguoYANG ChaoYANG Chao TYANG FuzhengYANG GuohuiYANG JinYANG PingYANG QinghuaYANG QinwenYANG WangYANG XiaolingYANG YanYANG YanhuiYANG YiYANG YongYANG ZhanminYAO YuanYAO ZanYAP Kim HuiYIN BoYIN Xun HeYOKOYAMA TakanoriYOON SeyoungYOSHIDA HaruyukiYOSHIDA MorioYOU KeyouYOU XianqiangYOUNG SOO SuhYU GangYU JianchengYU XinghuoYU Ze-FengYU ZhenyuYUAN ChangliangYUAN HuiYUAN JianYUAN JiaxinYUAN JingqiYUE Haosong

ZZAVERI MukeshZENG ChaoZENGER KaiZHAI LiZHAI Lian YinZHAI Xiaojun

136

ZHANG DamingZHANG DanhongZHANG DingguoZHANG Hai-TaoZHANG JingZHANG KeZHANG LZHANG LeiZHANG LijunZHANG QianniZHANG QinZHANG WeiZHANG WeizhongZHANG Wenlin

ZHANG XianxiaZHANG XiaolingZHANG YifanZHANG YingZHANG YuZHANG ZhiqiangZHANG ZhishengZHAO LijunZHAO WeiruiZHAO XiaoguangZHAO Yi ZhiZHEN ZiyangZHENG JianghuaZHENG Qi

ZHENG TaoZHIGUANG FengZHONG NingfanZHOU HuiyuZHOU JianguoZHOU QiZHOU RongZHOU SuipingZHU GZHU JiaqiangZHU XiaoruiZHUANG LiqunZOU YiqunZOU Yuanyuan

137

AUTHOR INDEX

Author Schedule PaperID

AA G Salazar-Gonzalez WA8.5 P0349ABASS Ahmed Fahmy TP3.5 P0673ABDEL-RAHEEM Esam WI.40 P0907ABDI Hamid WI.20 P0621ABDULLAH Azizi TM6.2 P0495ABDULLAH Soran Jalal TA6.1 P0254ACHTELIK Markus TI.13 P0517ACOSTA CALDERON C A FM5.1 P0842ADAM Alexander TM6.1 P0351ADAMS Martin David WM6.5 P0908AHMAD Tohari WA1.2 P1124AHMADI Majid WI.40 P0907AHMED Mousumi FM1.6 P1064AHMED Salim FP5.1 P0687AKBARIMAJD Adel WA7.1 P0509AKBARIMAJD Adel WA7.4 P0717AKMELIAWATI Rini TP5.1 P1089ALARFAJ Mishari FP7.2 P0515ALBERDI Mikel WA3.5 P0425ALBERTOS Pedro FM2.6 P1131AL-BLUWI Ibrahim TA3.1 P0342ALBOUL Lyuba FP6.3 P1046ALBRECHT Thomas FM6.3 P0844ALI Fariz WA7.2 P0584ALI Irshad FM6.1 P0702AL-KAYIEM Hussain H FI.11 P0488ALTHOEFER Kaspar WA6.4 P1060ALTHOEFER Kaspar WM6.2 P0357ALTOWATI Ali WM4.2 P0514AMELI Sina FP3.5 P0743AMJADIFARD Roya WM2.5 P0926AMUNDARAIN Modesto WA3.5 P0425AN Su-Yong WM7.1 P0320ANAZAWA Yoshihisa WA3.1 P0689ANDERSEN Palle FM1.3 P0824ANDERSON Tom A F FI.21 P0961ANG JR Marcelo H FP6.5 P1091ANG Lay Kee TA6.5 P0587ANVAR Amir WI.23 P0577APKARIAN Pierre TA1.2 P0225ARGENTIERI Sylvain FP6.2 P0992ARNUTTINANON B TA4.6 P0754ASADPOUR Masoud WA7.1 P0509ATYABI Adham FI.21 P0961ATYABI Adham WP5.5 P0951AUBRUN Christophe TP3.6 P1058AVANZINI Pierre FA5.2 P1120AVINASH H TA7.6 P1127AYCARD Olivier FI.20 P1085AYCARD Olivier WI.42 P0667AZMAN Amelia Wong FM6.4 P0852BBAEK Sang-Hoon WP6.3 P0347BAEK Young-Sik TI.22 P0537BAENFER Oliver WA2.2 P0632BAI Zhengyao FA6.3 P0171BAIG Mirza Salman FI.19 P0890BAIG Qadeer FI.20 P1085BAIG Qadeer WI.42 P0667

BALAMURALIDHAR P WI.32 P0405BARAMBONES Oscar WA3.5 P0425BARZAMINI Roohollah FI.1 P0378BASIM ISMAIL Firas FI.11 P0488BEKOOY Nico TA7.2 P1109BELIKOV Juri FP3.2 P0455BENN Wolfgang TM6.1 P0351BENSAALI Faycal TA7.1 P1107BHABA P K WA4.5 P0513BHATIA Divya WM3.6 P1071BHATTACHARJEE Arpita FP5.3 P0614BHATTI Asim FA3.1 P0308BHATTI Asim FA3.5 P0528BI Min TA4.1 P0871BIGDELI Abbas FM6.2 P0946BIGDELI Abbas FM6.4 P0852BIGDELI Abbas FP2.5 P1090BISCHOF Horst TA3.2 P0402BISCHOF Horst WA6.5 P0356BO Yuming WI.2 P0340BONNIFAIT Phillipe FA5.3 P1150BOTMART Thongchai FM2.2 P0438BOUSSARD Clément TP6.4 P0684BRIGHT Glen WP5.1 P0265BUI Tien Dai WI.41 P1063BURDAKOV Oleg TA5.4 P1155BURDET Etienne FM5.2 P0981BURLET Julien FI.20 P1085

CCAELLI Terry WI.35 P0872CAI Wenchuan WP6.4 P0780CAI Wen-Jian TI.1 P0247CAI Wen-Jian WI.12 P0835CAI Wen-Jian WM2.1 P0197CAI Wentong TM2.2 P0619CAI Wentong WM5.3 P0737CAI Yu-Jyun WM5.5 P0484CALDWELL Darwin G FP7.6 P0897CAO Feilong FA1.4 P0231CAPPELLI Raffaele WA1.4 P1146CAT Pham Thuong WI.6 P0982CAVEDON Lawrence FM5.4 P1042CERMAN Otto FI.9 P0142CHAI Hin Fei WA4.2 P0994CHAI Tianyou TM2.6 P0847CHAIVIVATRAKUL S TM2.4 P0731CHALIDABHONGSE T H FI.23 P1055CHAN Hian-Leng TP4.5 P1115CHANDRA M Girish TA7.6 P1127CHANDRA M Girish WI.32 P0405CHANDRA Pravin FP3.3 P1008CHANDRASEKAR S WA2.5 P0979CHANDWANI Vijay Kumar TI.15 P0682CHANG Ho-Chung FM3.4 P0680CHANG Kuochu TP1.3 P0811CHANG Wen-Chung TP7.4 P0611CHANG Yeong-Hwa TA5.2 P1152CHAPUIS Roland TM6.3 P1045CHAPUIS Roland WM7.3 P0785CHAUBEY Priyanka WP3.5 P0801

138

CHE AMRAN Aliza WA7.2 P0584CHE Wei Wei FI.5 P0986CHEAH Chien Chern FP1.2 P0967CHEAH Chien Chern WM6.6 P0984CHECCHIN Paul WM7.3 P0785CHEDED Lahouari FI.19 P0890CHELLAPPA Rama WA1.1 P1143CHEN Benmei WP4.3 P0483CHEN Chieh-Li FA7.3 P0879CHEN Chih-Chung TP7.6 P0913CHEN Chun-Lin TA5.2 P1152CHEN Dan WI.10 P0558CHEN Fengmin FP5.4 P0764CHEN Fengmin FP5.5 P0765CHEN Huiyan TM6.5 P0540CHEN Jiming WP4.1 P0403CHEN Lanping FI.2 P0304CHEN Liguo FP2.2 P0431CHEN Lingling FP6.1 P0894CHEN Peter C Y TA4.2 P1049CHEN Peter C Y WM5.6 P1026CHEN Qian TA6.4 P0346CHEN Weihai FA3.4 P0492CHEN Weiping FA6.5 P0812CHEN Wenjie FA3.4 P0492CHEN Wenjie FA3.6 P0736CHEN Wenjie WA5.1 P0656CHEN XiaoQi WI.20 P0621CHEN Yan TA4.1 P0871CHEN Zenghai FA7.1 P0490CHEN Zhiyong WM3.1 P0210CHENG Peng WP4.1 P0403CHENG Shan FA2.1 P0371CHENG Teddy M FI.18 P0324CHEONG Joono WI.24 P0896CHERFAOUI V´eronique FA5.3 P1150CHETOUANI Mohamed WM8.5 P0969CHI Zheru FA7.1 P0490CHI Zheru TI.25 P0609CHIANG Tsung-Che FM4.4 P0917CHLIVEROS Georgios FP6.3 P1046CHO Shung Han TI.16 P0690CHO Siu Yeung WP8.2 P1106CHO Siu Yeung WP8.3 P1108CHOI Han-Lim TP1.2 P0799CHOI Sang-Bok WA8.3 P0944CHOI Sung Eun TP2.4 P0719CHOI Young-Ho WP6.3 P0347CHONG Belinda C M TI.4 P0572CHONG Peter Han Joo WA4.2 P0994CHOUDHRY Muhammad WI.11 P0753CHOW J H WM2.6 P1105CHOW Tommy W. S. WP8.2 P1106CHU Delin WM4.1 P0422CHU Ping-Rung TP7.4 P0611CHUA Xing Lun WP7.4 P0991CIGLER Jiri WI.14 P0884COLLINS John James FM3.2 P0535CONG Nguyen Huu WI.8 P1053CORRALES Juan Antonio TI.19 P0413CREIGHTON Doug FA3.5 P0528CREIGHTON Doug FM4.6 P0831CREIGHTON Doug WM8.1 P0691CUI Meng WI.39 P0638CUONG Nguyen Huu FI.26 P0482

DDA Shi FA2.6 P0467DADGOSTAR Farhad FM6.4 P0852DAI Jian S WA6.3 P1018DAI Jing WI.10 P0558DAILEY Matthew N FM6.1 P0702DAILEY Matthew N TM2.4 P0731D'ANDRÉA-NOVEL Brigitte TP6.4 P0684DANG Ngoc Minh FI.31 P1102DARSIVAN Fadly Jashi TP5.1 P1089DASHTI J Mohammad FA4.2 P0714DE CHARETTE Raoul TP6.3 P0655DE GREGORIO Massimo FI.14 P1067DE LA SEN Manuel WA3.5 P0425DE LA SEN Manuel WM5.1 P0106DE MADRID Angel P TI.5 P0669DE VRIES Erik FM1.4 P1024DEBAIN Chistophe TM6.3 P1045DECRAENE James WM5.3 P0737DEHGHANI Maryam FA1.2 P1029DELMAS Pierre TM6.3 P1045DEMIRIS Yiannis WM8.2 P0539DENG Haibo WP5.4 P0602DEVLIN John WI.30 P1157DEY Rajeeb WP3.5 P0801DHARMAWEERA M N TP5.3 P1118DI STEFANO Luigi FA7.5 P0188DI STEFANO Luigi TM6.6 P0895DIEDERICH Joachim TA2.5 P0978DING Guoliang FP4.3 P0471DISSANAYAKE Gamini FM5.3 P0993DISSANAYAKE Gamini FP6.4 P0958DISSANAYAKE Gamini TA3.5 P1010DISSANAYAKE Gamini WM7.4 P0883DOHERTY Patrick TA5.3 P1154DOHERTY Patrick TA5.4 P1155DOHERTY Patrick TA5.5 P1156DOKI Kae TA2.2 P0427DOKI Shinji TA2.2 P0427DONG Jie FI.3 P0385DONG Jie WI.31 P0394DONG Li FI.24 P1070DONG Liang WP3.3 P0600DRISS Boutat TI.10 P0639DU Chunling WI.12 P0835DU Eliza Yingzi WP7.3 P1147DU Liqiang WI.15 P0960DUAN Suolin FI.2 P0304DUMUR Didier FM1.5 P1032DUNNIGAN Matthew W WM4.4 P0904DUNNIGAN Matthew W WM4.5 P0905DUTRA Max S FI.14 P1067DUY Nguyen Tien WI.8 P1053

EEFTAKHAR S. M. ASHIK FP2.6 P0659EKPANYAPONG Mongkol TM2.4 P0731EL HAMZAOUI Oussama FM3.3 P0660EL HAMZAOUI Oussama FM3.5 P1038ELFES Alberto FI.15 P0125ELNAGAR Ashraf TA3.1 P0342ELOUARDI Abdelhafid WM7.5 P0889EM Poh Ping TA2.4 P0505EMAMI M. Reza WM8.6 P1065

139

ER Meng Joo FM6.5 P0329ER Meng Joo FP3.1 P0983ER Meng Joo FP3.6 P0447ER Meng Joo TP4.6 P1128ER Meng Joo WA4.3 P1135ER Meng Joo WI.38 P0399ERIS Oytun WI.19 P0891ESLAMI Mehrad TI.26 P0790

FFAEZ Karim FA7.4 P0916FAEZ Karim TI.26 P0790FAN Huijin WM4.6 P0876FAN Wei TI.21 P0943FAN Yuan WM3.2 P0249FANG Huajing WI.17 P0696FANG Huajing WM3.4 P0500FANG Jianyin TI.9 P0565FANG Qingxiang FA1.4 P0231FARIS Waleed F FI.10 P0333FARZINFAR Mahshid TM2.5 P0755FATEHI Nima WA7.1 P0509FENG Dagan FA7.1 P0490FENG Dagan TI.25 P0609FENG Gang WM3.2 P0249FENG Jun-e WP3.4 P0966FENG Zhi-Yong WA3.1 P0689FERRARA Matteo WA1.4 P1146FOUDIL Abdessemed TI.10 P0639FOURQUET Jean-Yves WM8.4 P0965FOWERS Spencer G FP2.3 P0623FRANÇA Felipe M G FI.14 P1067FRANÇA Horacio L FI.14 P1067FRAUNDORFER Friedrich TI.13 P0517FU Hong FA7.1 P0490FU Yongling WI.13 P0874FUCHS Julian TI.3 P0453FUJIOKA Hisaya TA1.3 P0603

GG BALAKRISHNAN TI.24 P0437GAN Oon Peen TP4.1 P0923GAN Oon Peen TP4.6 P1128GANI Mahbub WI.21 P0526GAO Daoxiang FM5.5 P0759GAO Deyun FI.13 P0498GAO Guoqin FM1.1 P0463GAO Junbin TA7.3 P1113GAO Tingting WI.12 P0835GAO Weimin WI.22 P0567GAO Xuehai TI.18 P0851GAO Yongsheng FA6.2 P0937GAO Yongsheng FA6.5 P0812GAO Yongsheng WI.35 P0872GAO Zhanfeng WI.15 P0960GARCIA DE MARINA H WP5.2 P0423GARCIA FAVROT Olivier TP6.1 P0502GARCIA Gabriel J TP7.1 P0396GARCIA GIL Pedro FM2.6 P1131GARRATT Matt TA1.6 P0368GARRIDO Aitor J. WA3.5 P0425GARRIDO Izaskun WA3.5 P0425GE Ruowen TA4.2 P1049GE Ruowen WM5.6 P1026

GEORGE Koshy WM4.3 P0816GERAMIFARD Omid TP4.4 P0977GHAHRAMANI M FI.31 P1102GHAHRAMANI M TI.31 P1001GHANBARI Ali WI.20 P0621GHOSH Sandip WP3.5 P0801GIRON-SIERRA Jose M WP5.2 P0423GLAZUNOV Victor FA4.3 P0828GO Tiauw Hiong WM3.3 P0282GOH Kah Ong Michael TP2.2 P1132GOH Kiah-Mok TP4.5 P1115GOH Wooi-Boon WA5.3 P0826GONG Jianwei TM6.5 P0540GONG Jun FP6.5 P1091GONG ZhingMing FI.25 P1104GOODMAN Erik TA6.2 P0281GOODWINE Bill TI.11 P1068GOPALAN Vani FM7.5 P0817GORDON FANE Anthony WM2.4 P0850GOURIVEAU Rafael TP4.1 P0923GOVERNATORI Guido TA2.5 P0978GRAVDAHL Jan Tommy WA7.3 P0267GRUYER Dominique WM7.5 P0889GU Guohua TA6.4 P0346GU Nong WI.22 P0567GUAN Zhenying WI.22 P0567GUERRA MORALES C TP6.6 P0520GULAM RAZUL S WP2.3 P0867GUNN Bruce FM4.6 P0831GUPTA Phalguni FA6.1 P0672GUPTA Raj Kumar WP8.3 P1108

HHA In-Joong TP7.5 P0688HA Le Thi Thu WI.8 P1053HA Q P WP3.1 P0381HÄGGBLOM Kurt E WP2.4 P1035HAGHIGHI Reza WM6.6 P0984HAJATI Farshid FA6.2 P0937HAKIMA Ali FP3.5 P0743HAN Bingxin WI.15 P0960HAN Lixia WM4.6 P0876HANIF Aamir WI.11 P0753HANN Christopher E FA3.3 P0360HANSEN Michael FM1.3 P0824HARSHANGI Prashanth WM4.3 P0816HARTMANN Benjamin WA2.2 P0632HARUN Mohd Hanif Bin TA2.4 P0505HASBULLAH Faried FI.10 P0333HASHIMOTO Kohjiro TA2.2 P0427HASHIZUME Hideki FP1.4 P1036HATZINAKOS Dimitrios FA6.3 P0171HAUGSTUEN Idar U FP7.1 P0268HAUTI`ERE Nicolas TP6.4 P0684HAYASAKA Kazuto WI.29 P0784HE Chun-yu WI.1 P0309HE Mingyi TI.30 P1130HE Xiangjian TA7.3 P1113HE Xiao FI.6 P0493HEDE Patrick TA2.6 P1054HEINTZ Fredrik TA5.3 P1154HERATH Damith C FM5.4 P1042HERNANDEZ M Angeles WI.7 P0361HERN´ANDEZ Luis TP6.6 P0520HIDANO Seira TP2.3 P1149

140

HIRCHE Sandra TP7.6 P0913HOE Jerry Kah Eng FI.24 P1070HOLMBERG Kaj TA5.4 P1155HONG Sangjin TI.16 P0690HONG Sangjin WI.28 P0692HONG Sangseok WA8.3 P0944HORAN Ben WI.20 P0621HOSSNY Mohammed FA3.1 P0308HOSSNY Mohammed FA3.5 P0528HOU Qian WM5.4 P0902HOU Saing Paul FP1.2 P0967HOU Zengguang WP5.4 P0602HOVER Franz S. WM6.5 P0908HOW Jonathan P. TP1.2 P0799HOY Michael TA1.4 P0370HOY Michael TA1.6 P0368HREDZAK Branislav FP4.1 P0833HSU Roy Chaoming TI.29 P1028HU Gibson WM7.4 P0883HU Jiankun WA1.2 P1124HU Jiankun WA1.3 P1137HU Wuhua TI.1 P0247HU Wuhua WM2.1 P0197HUA Jianning FA5.5 P0352HUANG Guangpu FM6.5 P0329HUANG Jung-Shou WI.33 P0151HUANG Sheng TP4.6 P1128HUANG Shoudong WM7.4 P0883HUANG Weifeng TP7.3 P0560HUANG Xiaosheng WI.10 P0558HUANG Xinsheng WI.37 P0372HUANG Xiwei FA7.2 P0769HUDHA Khisbullah TA2.4 P0505HUNEK Wojciech P TP3.3 P1020HUSEK Petr FI.9 P0142HUSEK Petr WA2.3 P0141HUSSAINI Safiullah WA5.3 P0826HWANG Yean-Ren WI.9 P0185

IIBEAS Asier WM5.1 P0106ICHINO Masatsugu WP7.5 P1148IGARASHI Tatsuo WA8.4 P0881IHARA Naoya FA2.5 P0248IMAM Saiyad A WI.4 P0671INADA Hiroshi WI.3 P0414INANC Tamer FI.15 P0125ISHII Ken WI.3 P0414ISHII Takuro WA8.4 P0881ISHIKAWA Masatoshi WP6.2 P1013ISHIKAWA Seiji FP2.6 P0659ISMAIL Napsiah FM3.1 P0421ISMAIL Zool H WM4.4 P0904ISMAIL Zool H WM4.5 P0905IU Herbert Ho-Ching TP3.4 P0935IZADINIA Hamid TM6.4 P1056

JJ Mary Anita TI.12 P0419JAIN Siddarth TI.15 P0682JAMAL Arif WI.4 P0671JAMAL Maria WI.4 P0671JAMALUDDIN Hishamuddin TA2.4 P0505JANA Amiya K TP5.6 P0263

JARA Carlos A TI.19 P0413JARA Carlos A TP7.1 P0396JARVIS Ray FI.27 P0601JAYASUNDARA Kosala FP4.4 P0604JEONG Jae Jin TI.14 P0610JI Yindong FI.6 P0493JIANG Bin TA6.6 P0976JIANG Bin WM5.4 P0902JIANG Daogen FM1.1 P0463JIAO Zhi-jie WI.1 P0309JIMENEZ Juan Francisco WP5.2 P0423JIMENEZ-LOZANO Joel TI.11 P1068JIN Yi FP4.3 P0471JOHNSON-ROBERSON M WM8.3 P0661JONKER Pieter P FP2.4 P0681JOSHI Maulin M WM6.3 P0553JUNG Ho Gi WP7.1 P0735

KKABILEN Sornum TM2.2 P0619KAEWKHAM-AI Boonsri WP2.5 P0321KAJIKAWA Shinya WI.29 P0784KALAM Shemin WI.21 P0526KALIA Meenakshi FI.29 P0114KALYAN Bharath TP6.5 P0770KAMATA Sei-ichiro WP8.5 P1114KANAYAMA Takafumi FP1.4 P1036KANG Hee Jun FA2.2 P0934KANG Jeong-Gwan WM7.1 P0320KANG Rui WI.18 P0793KANTOR George FP7.2 P0515KAO Ping-Wen TI.29 P1028KASHYAB Pushpinder FI.8 P1095KAWAMURA Atsuo WA7.2 P0584KAWAMURA Sadao FP1.4 P1036KEALL Alex FA3.3 P0360KELLER Samuel WM6.5 P0908KESHMIRI Mehdi FP7.3 P0658KHALID Muhammad FP4.6 P0955KHALOOZADEH Hamid WM2.5 P0926KHAN Asad I. FM7.6 P0945KHAN M R WI.4 P0671KHAN Masood Mehmood WI.34 P0266KHOO Li Pheng WM2.6 P1105KHOO Suiyang TP5.3 P1118KHOO Suiyang TP5.4 P1145KHOSRAVI Abbas FM4.6 P0831KIA Solmaz WA7.4 P0717KIM Daehyun TI.14 P0610KIM Daijin FM6.6 P1142KIM Gahyun WP7.1 P0735KIM Hakil WP7.2 P1121KIM Hyoungseop FP2.6 P0659KIM Jaihie TP2.4 P0719KIM Jaihie WP7.1 P0735KIM Jungho FM7.2 P0571KIM Kiseon WA3.2 P0706KIM Sang-woo TI.14 P0610KIM Sungho FM7.2 P0571KINSHEEL Azeddien WA6.1 P0704KLEINJOHANN Bernd WM7.2 P0475KLEINJOHANN Lisa WM7.2 P0475KODAGODA Sarath FM5.3 P0993KODAGODA Sarath FP6.4 P0958KOMATSU Naohisa TP2.3 P1149

141

KOMATSU Naohisa WP7.5 P1148KONG Jianshou WI.2 P0340KONIGORSKI Ulrich TI.3 P0453KOREVAAR Agnetha FA3.3 P0360KRISHNA K. Madhava TA3.4 P0677KROLL Andreas WP6.5 P0272KROOS Christian FM5.4 P1042KU Chang-Ping WA8.4 P0881KÜHNLENZ Kolja TP7.2 P0507KÜHNLENZ Kolja TP7.6 P0913KULIC Dana WP4.4 P0825KVARNSTRÖM Jonas TA5.4 P1155KVARNSTRÖM Jonas TA5.5 P1156KWEON In So FM7.2 P0571KWON Jeong-Hun TP7.5 P0688

LLADEVEZE Nicolas WM8.4 P0965LAI Chow Yin FP5.2 P0554LAI Wei-Jie TI.29 P1028LAM Siew Kei TI.28 P0845LAMBERT Alain WM7.5 P0889LATAWIEC Krzysztof J TP3.3 P1020LAUGIER Christian TP6.2 P0511LE TRAN Hoai Nam TM2.2 P0619LE Vu Thanh WM8.1 P0691LEE Chong Yee FA6.6 P0586LEE Dah-Jye FP2.3 P0623LEE Dah-Jye TM6.5 P0540LEE Gim Hee TI.13 P0517LEE Je Won TI.14 P0610LEE Jinseok WI.28 P0692LEE JinSoo WA8.3 P0944LEE Keon-Ho TM2.3 P0713LEE Lae-Kyoung WM7.1 P0320LEE Lay-Lan TI.32 P0407LEE Sang-Deok TI.22 P0537LEE Sung Joo TP2.4 P0719LEE Tae-Kyeong WP6.3 P0347LEE Tong Heng FP5.2 P0554LEE Youn Joo TP2.4 P0719LEES Michael Harold TM2.2 P0619LEFRANC Pierre FM1.5 P1032LENG Yu-Chi FA7.3 P0879LENGERKE Omar FI.14 P1067LENZ Martin WA6.5 P0356LEOPUTRA Wilson S FP2.1 P0212LEROUX Christophe TA2.6 P1054LEUOTH Sebastian TM6.1 P0351LI Ai-Jun FM1.2 P0258LI Cai WA1.3 P1137LI Cao FM4.3 P0901LI Daoping FM4.2 P0470LI Dawei TA6.2 P0281LI Dequan WM3.5 P0529LI Guoqi WP2.2 P0760LI Honghai FP3.4 P0534LI Hongyi FA5.5 P0352LI Hui FP4.5 P0925LI Jianxun FA2.3 P0987LI Jun FA6.6 P0586LI Lin WP3.3 P0600LI Liyuan FI.24 P1070LI Long TI.30 P1130LI Ning TI.2 P0389

LI Ping TA4.3 P0409LI Pu WP2.1 P0616LI Qian TI.2 P0389LI Shao-yuan TI.2 P0389LI Wanqing FA6.4 P0605LI Wanqing TA7.4 P0429LI Weitao TM2.6 P0847LI Xiang FP1.2 P0967LI Xiang FP3.6 P0447LI Xiang TA6.5 P0587LI Xiang TP4.1 P0923LI Xiang TP4.4 P0977LI Xiang TP4.6 P1128LI Xin TM6.5 P0540LI Xingwei WI.37 P0372LI Xingxiu WI.2 P0340LI Yangmin FI.16 P0940LI Yan-Ping FI.5 P0986LI Yan-Ping WP3.3 P0600LI Yong-Fu FA1.6 P0757LI Yongmin WA8.5 P0349LI Zhenning WP4.4 P0825LI Zhi FI.27 P0601LI Zhibin FP7.6 P0897LIA Alexandre TP6.3 P0655LIAN Guofu FP4.3 P0471LIAN Zhichao WI.38 P0399LIANG Bin TI.18 P0851LIANG Wenfeng FP5.6 P0857LILJEBÄCK Pål FP7.1 P0268LILJEBÄCK Pål WA7.3 P0267LIM Beng Siong TP4.6 P1128LIM Chin-heng WP2.3 P0867LIM Hock Beng WA4.3 P1135LIM Meng-Hui TP2.5 P0726LIM Ooi Chong TI.17 P0782LIM Teck-Yian TP4.5 P1115LIN Hai WP4.3 P0483LIN Shin-Yeu WI.33 P0151LIN Wei FA3.6 P0736LIN Wei TA4.2 P1049LIN Wei WM2.6 P1105LIN Wei WM5.6 P1026LIN Wen-Jong FI.25 P1104LIN Wen-Jong WM2.6 P1105LIN Zhiping WA3.4 P1129LIN Zhiping WP2.3 P0867LING Keck Voon TP4.5 P1115LING Keck Voon WM2.4 P0850LIU Cheng-Ting TI.29 P1028LIU Chien-Wei TI.32 P0407LIU Chongjie FA5.4 P0523LIU Chongjie WI.27 P1073LIU Fan FP3.1 P0983LIU Feng WM4.1 P0422LIU Gang TA1.5 P0323LIU Hao TI.21 P0943LIU Hong FM4.3 P0901LIU Kai TI.9 P0565LIU Lixian WI.15 P0960LIU Ning TA6.4 P0346LIU Shuai TA5.2 P1152LIU Wei WA4.1 P0985LIU Wei WA4.2 P0994LIU Wenwei WP3.3 P0600LIU Xiaohui WA8.5 P0349

142

LIU Xiaomeng WP4.3 P0483LIU Xinjun FM1.1 P0463LIU Xinzhi FM2.2 P0438LIU Yi-Hung WI.36 P0179LIU Zhitao WP2.1 P0616LOH Ai Poh TA7.5 P0548LOU Lei TP7.6 P0913LOVELL Brian C FM6.2 P0946LOVELL Brian C FM6.4 P0852LOW Malcolm Yoke Hean TM2.2 P0619LOW Malcolm Yoke Hean WM5.3 P0737LOW Tobias TA3.3 P0406LU Chia-Hua WI.9 P0185LU Dunmin FM5.5 P0759LU Guirong FI.2 P0304LU Jianhua WM5.4 P0902LU Jiwen WI.39 P0638LU Ningyun WM5.4 P0902LU Tien-Fu WI.23 P0577LU Tien-Fu WM5.2 P0245LU Zhe TA4.2 P1049LU Zhe WM5.6 P1026LUCA Anamaria FM1.5 P1032LUKANISZYN Marian TP3.3 P1020LUM Kai Yew TA1.1 P0989LUO Haibo WA8.1 P0224LUO Hong FI.25 P1104LUO Hong TA4.2 P1049LUO Hong WM5.6 P1026LUO Ming TP4.3 P0962LUO Ningsu WM6.1 P0227LUO Yan FM1.1 P0463LUO Zhizeng FM7.4 P0665LUU Khoa WI.41 P1063LY Thanh FM6.3 P0844

MM Behnam T FA4.2 P0714M Benazir FI.30 P0882M D P Moratuwage TP6.5 P0770M D P Moratuwage WP6.1 P0300M Malleswaran FI.12 P0418M Malleswaran TI.12 P0419M Poorani FP6.6 P0472M Sankari FI.30 P0882M Vijayakarthick WA4.5 P0513M WILLJUICE I FI.7 P1088MA DI WA4.3 P1135MA Li-Shan WM5.5 P0484MA Xueming WI.31 P0394MA Zhenghua FI.2 P0304MAHINDRAKAR Arun D FA1.1 P0147MAHINDRAKAR Arun D FP4.2 P0146MAINETTI Andrea FA7.5 P0188MAIRE Frederic FP2.5 P1090MAJDI Laila WA7.1 P0509MAJHI Somanath TI.7 P0709MAKINO Harufumi WA8.4 P0881MAKUR Anamitra WA3.4 P1129MALARTRE Florent TM6.3 P1045MALASHETTY Ravi S WA5.5 P0800MALIK Arshad WI.11 P0753MALTONI Davide WA1.4 P1146MAN Zhihong TP5.3 P1118MAN Zhihong TP5.4 P1145MANOSO Carolina TI.5 P0669

MANVI Sunilkumar S WA5.5 P0800MANZANERA Antoine TA3.3 P0406MAO Kezhi TM2.6 P0847MAO Pengxuan WA3.2 P0706MAO Yong TP6.2 P0511MAREELS Iven FM5.2 P0981MARTIN Adrian WM8.6 P1065MARTINET Philippe FA5.2 P1120MART´INEZ Alain TP6.6 P0520MARTONO Wahyudi TP5.1 P1089MASSOL Olivier TP4.1 P0923MASSOL Olivier TP4.6 P1128MATTOCCIA Stefano FA7.5 P0188MATTOCCIA Stefano TM6.6 P0895MATVEEV Alexey S TA1.4 P0370MAYER Heinz TA3.2 P0402MEHMOOD Tahir WI.11 P0753MEHTA Utkal TI.7 P0709MENDEZ Gerardo M. WI.7 P0361MENG Ming FM7.4 P0665MENG Wen-Jie FM1.2 P0258MENIGHED Kamel TP3.6 P1058MIAO Yuan WA5.2 P0744MIKI Tomohiro WI.3 P0414MOGHADAM Peyman TP6.5 P0770MOGHADAM Peyman WP6.1 P0300MOHAMMAD AMIN A H FM7.6 P0945MOHAN M Agnes Saranya WM7.6 P0615MOHAN Mahesh TA3.4 P0677MOHAN Rajesh Elara FM5.1 P0842MOHD MUSTAFAH Yasir FM6.4 P0852MOHD THAN Mohd Nor TI.4 P0572MOONRINTA Jednipat TM2.4 P0731MORAS Julien FA5.3 P1150MORIGNOT Philippe TA2.6 P1054MOTLAGH O FM3.1 P0421MULLANE John WM6.5 P0908MURAKAMI Shigeo FP1.4 P1036MUSOROMY Zuwena TA7.2 P1109MUTLU Ilhan WI.19 P0891MYO Maung Thi Rein WM5.2 P0245MYSOREWALA M F FI.19 P0890

NN Narayanaswamy WA2.5 P0979N T Naresh Babu TA7.6 P1127NABATCHIAN Amirhosein WI.40 P0907NABI Masque Un FI.8 P1095NAHAVANDI Saeid FA3.1 P0308NAHAVANDI Saeid FA3.5 P0528NAHAVANDI Saeid FM4.6 P0831NAHAVANDI Saeid WI.20 P0621NAHAVANDI Saeid WI.22 P0567NAHAVANDI Saeid WM8.1 P0691NAM Joo Hoo TA4.2 P1049NAM Joo Hoo WM5.6 P1026NAM Yunyoung WI.28 P0692N Sundararajan FM7.5 P0817N Sundararajan WA2.5 P0979NARUMI Makoto WI.3 P0414NASHASHIBI Fawzi TP6.1 P0502NASHASHIBI Fawzi TP6.3 P0655NASIR Adnan TP1.4 P1117NASUNO Masashi WI.29 P0784NAYA Yukio WA8.4 P0881

143

NEGRE Amaury TP6.2 P0511NELLES Oliver WA2.2 P0632NG Chuen Leong FI.25 P1104NG Chuen Leong WA5.1 P0656NG Teck Chew WA5.1 P0656NGUWI Yok Yen WP8.1 P1110NGUWI Yok Yen WP8.4 P1111NGUYEN Duc Thanh TA7.4 P0429NGUYEN Hung T WI.16 P1080NGUYEN Hung T WP3.1 P0381NGUYEN Minh Thanh FA4.3 P0828NGUYEN Thanh Van Thi FA5.6 P0516NGUYEN Thi Anh Ngoc WA5.1 P0656NGUYEN Thi Hai Binh WP7.2 P1121NGUYEN Van Huan WP7.2 P1121NGUYEN Xuan Vinh FA4.3 P0828NIA D. Nakhaei FM3.1 P0421NIAMSUP Piyapong FM2.2 P0438NIELSEN K M FM1.3 P0824NIGAM Aditya FA6.1 P0672NIKRAVESH S K FA1.2 P1029NILKUND Prashant S FA4.6 P0906NISHIO Kimihiro FA2.5 P0248NOROUZNEZHAD Ehsan FM6.2 P0946NQUYEN-DUC Hiep TA2.3 P0428

OOETOMO Denny FM5.2 P0981OGUNBONA Philip FA6.4 P0605OGUNBONA Philip TA7.4 P0429OH Se-Young WM7.1 P0320OH Se-Young WP6.3 P0347OHKI Tetsushi TP2.3 P1149OKUMA Shigeru TA2.2 P0427OLSSON Per-Magnus TA5.4 P1155ONG Chong Jin WM5.6 P1026ONG Shen-Hoong TP4.5 P1115ONG Sim Heng TI.33 P0580ONISHI Hiroyuki FP1.4 P1036ONUNKA Chiemela WP5.1 P0265OOI Kelvin Jian Aun TA6.5 P0587OONCHOM Keelati TA4.6 P0754OSOTHSILP Anan TP7.3 P0560

PP Balmuralidhar TA7.6 P1127P HA Quang TA2.3 P0428P HA Quang WI.16 P1080P T V Bhuvaneswari WI.32 P0405P T V Bhuvaneswari WM7.6 P0615PACHECO Lluís WM6.1 P0227PAI M M Manohara WI.26 P0748PAN Shuwen WP2.1 P0616PANDI V R FA1.3 P1096PANDIAN Shunmugham R FP1.3 P1000PANG Wee Ching WP5.3 P0589PANIGRAHI B. K. FA1.3 P1096PARADOWSKI Mariusz TM2.1 P0415PARENT Michel TP6.1 P0502PARK Adrian WA8.2 P0918PARK Chaehoon FM7.2 P0571PARK Do-Hwan TP7.5 P0688PARK Ji-Ho TI.22 P0537PARK Kang Ryoung TP2.4 P0719

PARK Tae-Hyoung TM2.3 P0713PARK Youngsu TI.14 P0610PAROMTCHIK Igor E TP6.2 P0511PATEL Vishal M WA1.1 P1143PATIL Manish M FA4.6 P0906PATRIKALAKIS N M TP6.5 P0770PATRIKALAKIS N M WM6.5 P0908PEDERSEN T S FM1.3 P0824PENG Guangzheng TA4.5 P0700PENG Kaixiang FI.3 P0385PENG Kaixiang WI.31 P0394PEREDA Fernando J WP5.2 P0423PERROLLAZ Mathias TP6.2 P0511PERUMAL Dananjayan FA2.4 P1023PERUMAL Dananjayan TA4.4 P0596PETLENKOV Eduard FP3.2 P0455PETTERSEN Kristin Y FP7.1 P0268PETTERSEN Kristin Y WA7.3 P0267PHAN Tuan A WM6.4 P0795PHUNG Manh Duong FA5.6 P0516PIRKER Katrin TA3.2 P0402PIZARRO Oscar WM8.3 P0661POLLEFEYS Marc TI.13 P0517POMARES Jorge TP7.1 P0396P N Suganthan FA1.3 P1096P N Suganthan FI.7 P1088P Pramod Kumar TA7.5 P0548POPA Dan O FI.19 P0890POSTULA Adam FM6.2 P0946POURBOGHRAT Farzad TP7.3 P0560POWERS David M W FI.21 P0961POWERS David M W WP5.5 P0951PREMARATNE Prashan FM5.4 P1042PRIVARA Samuel WI.14 P0884PROCHAZKA Zdenek TI.27 P0832

QQIAO Hanping TA2.1 P0915QIN Yajuan FI.13 P0498QU Bo-Yang FA1.3 P1096QU Bo-Yang FI.7 P1088QU Guangzhi WA3.2 P0706QUACH Cong Hoang FA5.6 P0516

RR Bhargavi WI.32 P0405R Bremananth FI.30 P0882R Sundaramurthy TA4.4 P0596R Vijayanandh TI.24 P0437RAHMAN M. K. M. WP8.2 P1106RAI Chandra Shekhar FP3.3 P1008RAIE Abolghasem A. FA6.2 P0937RAJKUMAR K Timothy FA2.4 P1023RAJPUROHIT Vijay S WI.26 P0748RAMACHANDRAN S TP1.4 P1117RAMACHANDRAN S WA4.1 P0985RAMALINGAM Shankar FA2.4 P1023RAMALINGAM Soodamani TA7.1 P1107RAMALINGAM Soodamani TA7.2 P1109RAMASAMY Savitha FM7.5 P0817RAMLI Abdul Rahman FM3.1 P0421RAO Akshay WM6.5 P0908RAO Avinash FA3.3 P0360RASCHE Christoph WM7.2 P0475

144

RATNAM Mani Maran TA6.1 P0254RAVICHANDRAN A A FP4.2 P0146RAY Goshaidas WP3.5 P0801REZVANI DASTGERDI H FP7.3 P0658RICANEK Karl WI.41 P1063RIISGAARD-JENSEN M FM1.3 P0824RO Young Shick FA2.2 P0934ROBBEL Philipp FM7.3 P0592RODR´IGUEZ Yidier TP6.6 P0520RODRIGUEZ-AYERBE P FM1.5 P1032ROHANI Fateme WM2.5 P0926ROMERO Miguel TI.5 P0669RONG Weibin FP2.2 P0431ROSS Robert WI.30 P1157ROSSMANN Juergen WA5.4 P0909ROY Balai Chandra TP5.5 P1047ROY Deb FM7.3 P0592RUAN Xiaogang FP7.5 P0724RUDINAC Maja FP2.4 P0681RUSSELL R. Andrew WM6.4 P0795RÜTHER Matthias TA3.2 P0402RÜTHER Matthias WA6.5 P0356

SS Angel Deborah FI.12 P0418S Baskar FI.7 P1088S CHAUDHARI Narendra FA4.1 P0624S CHAUDHARI Narendra FM4.5 P0620S J Mija TP3.1 P0380S Janardhanan TP3.2 P0386S Manjula FI.12 P0418SADEGHI Fereshteh TM6.4 P1056SAEEDI Parvaneh FM7.1 P0159SAEI MANESH Fahimeh TI.31 P1001SAFABAKHSH Reza TM6.4 P1056SAHLI Hichem TP6.6 P0520SAKAGAMI Norimitsu FP1.3 P1000SAKAGAMI Norimitsu FP1.4 P1036SAKURAI Yasunori WI.3 P0414SAMAD Zahurin TA6.1 P0254SAMIEI Ehsan WA6.2 P1012SAMMUT Claude TI.23 P0306SAN Linn FP3.6 P0447SANGPRADIT Kiattisak WA6.4 P1060SANTHANAM Sathishbabu WA4.5 P0513SATYANARAYANA Neeli TP3.2 P0386SAVKIN Andrey V FI.18 P0324SAVKIN Andrey V FP4.6 P0955SAVKIN Andrey V TA1.4 P0370SAVKIN Andrey V TA1.6 P0368SAWLANI Manish TI.15 P0682SCHWEIGHOFER Gerald TA3.2 P0402SEAH Hock Soon TM2.2 P0619SEE Chong-Meng WP2.3 P0867SEET Gerald WP5.3 P0589SEHESTEDT Stephan FM5.3 P0993SEIGNEZ Emmanuel WM7.5 P0889SENEVIRATNE Lakmal D WA6.4 P1060SENEVIRATNE Lakmal D WM6.2 P0357SENGUPTA Anindita FP5.3 P0614SENOO Taku WP6.2 P1013SEVILLANO Jose M Goretti WA3.5 P0425SEYEDKASHI S M H FA4.2 P0714SHAFIEE Masoud FI.1 P0378SHAFIEE Masoud WA6.2 P1012

SHE Qingshan FM7.4 P0665SHEKHAR Raj WA8.2 P0918SHENTON Andrew Thomas TP3.5 P0673SHI Guodong WM4.1 P0422SHI Lei FP6.4 P0958SHI Ling WP4.1 P0403SHI Ling WP4.5 P1069SHIBATA Mizuho FP1.4 P1036SHIH Ta-ming FM3.4 P0680SHIH Ta-ming TI.8 P0541SHIRINZADEH Bijan WI.25 P0532SHIROKU Reyes Tatsuru FP1.1 P0959SHOJAEE G Kambiz FA4.2 P0714SHTESSEL Yuri TA1.5 P0323SIDNAL Nandini S WA5.5 P0800SIEGWART Roland TI.13 P0517SILVA João Carlos FI.14 P1067SIM Siang Kok WP5.3 P0589SINGH Amit Prakash FP3.3 P1008SINHA Nishant WP3.5 P0801SKINNER Greg FA3.3 P0360SLUZEK Andrzej TM2.1 P0415SN Sabreen TI.12 P0419SNOWDON Malcolm FA3.3 P0360SOH Chin Yun TI.17 P0782SOH Yeng Chai FI.4 P0775SOLDAN Samuel WP6.5 P0272SONG Insu TA2.5 P0978SONG Qi FA4.4 P0869SONG Qing FI.17 P1158SONG Xiaojing WM6.2 P0357SONG Xinmin WP4.5 P1069SONG Yongduan FA4.4 P0869SONG Yongduan WP6.4 P0780SOOD Ridhi FI.29 P0114SOONG Boon-Hee TP1.4 P1117SOONG Boon-Hee WA4.1 P0985SORAYA Bououden TI.10 P0639SOURY Mariette TA2.6 P1054SREERAM Victor TP3.4 P0935SRIDHARAN K TA3.6 P1103SRIKANTHAN Thambipillai TI.28 P0845STAVDAHL Øyvind WA7.3 P0267STERN Claudius WM7.2 P0475STEUX Bruno FM3.3 P0660STEUX Bruno FM3.5 P1038STEVENS Catherine J FM5.4 P1042SU Chong-Wei WI.33 P0151SU Hongye WP2.1 P0616SU Steven Weidong WI.16 P1080SUBBARAMAN Shaila FA4.6 P0906SUBBARAO Kamesh FM1.4 P1024SUBBARAO Kamesh FM1.6 P1064SUBBARAO Kamesh WM3.6 P1071SUDSANG Attawith TI.20 P0512SUEN Ching Y WI.41 P1063SUH Young Soo FA2.2 P0934SUHR Jae Kyu WP7.1 P0735SUI Xiubao TA6.4 P0346ŠULC Bohumil TI.6 P1037SUMITRA A TA7.6 P1127SUN Di-Hua FA1.6 P0757SUN Dong WA7.5 P0445SUN Hao FP6.1 P0894SUN Jitao FM2.3 P0477SUN Kui FM4.3 P0901

145

SUN Lin WP3.4 P0966SUN Lining FP2.2 P0431SUN Wei TP1.3 P0811SUN Ying TI.33 P0580SURESH Sundaram FI.8 P1095SURESH Sundaram WA2.5 P0979SUSHKOV Oleg O TI.23 P0306SUTRADHAR Ashoke FP5.3 P0614SUTTON Erica WA8.2 P0918

TT R Maruthi FA1.1 P0147TAHA Zahari WA6.1 P0704TAHERI ASHTIYANI F FA7.4 P0916TAI Zhixin WP3.2 P0156TAKAHASHI Kenta TP2.3 P1149TAKAYAMA Go WI.3 P0414TAKEMURA Fumiaki FP1.1 P0959TAMMI Kari WM4.2 P0514TAN Jian FM1.2 P0258TAN Joo Kooi FP2.6 P0659TAN Kuan Meng WI.23 P0577TAN Shaohua FA2.6 P0467TAN Tele FM6.3 P0844TAN Tele FP2.1 P0212TAN Yap-Peng TI.31 P1001TAN Yap-Peng WI.39 P0638TAN Yap-Peng WP7.4 P0991TANG Robert FA3.3 P0360TANG Sai Hong FM3.1 P0421TANG Yuangui FA5.4 P0523TANG Yuangui WI.27 P1073TANNO Yuichi WP6.2 P1013TAO Guocai TA4.1 P0871TAPUS Adriana WM8.5 P0969TAY Bertrand WA4.2 P0994TAY Christopher TP6.2 P0511TEACHABARIKITI Kosit FI.23 P1055TEE Connie TP2.2 P1132TENNAKOON Salinda FP4.4 P0604TEOH Andrew Beng Jin TP2.2 P1132TEOH Beng Jin Andrew TP2.5 P0726TEOH Eam Khwang FI.31 P1102TEOH Eam Khwang TM2.5 P0755TEOH Teik Toe WP8.1 P1110TEOH Teik Toe WP8.4 P1111THAI HOANG Huynh FI.26 P0482THAI Vu Duc WI.6 P0982THAM Chen Khong FP6.5 P1091THAM Ronny Quin Fai FP6.5 P1091THAMMANO Arit FI.23 P1055THOMAS Jaya FA4.1 P0624THOMAS N. Luke WP7.3 P117THOMAS Susy TP3.1 P0380THONNAGITH Peerapong TI.20 P0512THUILOT Benoit FA5.2 P1120TIAN Yanbing TA4.5 P0700TIAN Yanling WI.25 P0532TING Yung WI.36 P0179TISTARELLI Massimo WA1.1 P1143TISTARELLI Massimo WA1.4 P1146TIWARI Aruna FA4.1 P0624TIWARI Aruna FM4.5 P0620TOH Kar-Ann TP2.1 P0798TOMBARI Federico FA7.5 P0188

TOMBARI Federico TM6.6 P0895TORABI JAHROMI Amin FP3.6 P0447TORABI JAHROMI Amin TP4.6 P1128TORII Akihiro TA2.2 P0427TORRES Fernando TI.19 P0413TORRES Fernando TP7.1 P0396TRAN Cong Tuan FA4.3 P0828TRAN Quang Vinh FA5.6 P0516TRAN Tri WM2.2 P0223TRAN Tri WP3.1 P0381TREHARNE Kenneth FI.21 P0961TRUNG Ngo Kien WI.8 P1053TSAGARAKIS Nikos G FP7.6 P0897TSAI Gwo-Chung TI.32 P0407TUAN Hoang Duong TA1.2 P0225

UUEDA Tomohiro FP1.4 P1036UTHAICHANA Kasemsak WP2.5 P0321

VVACHHANI Leena TA3.6 P1103VADAKKEPAT Prahlad TA7.5 P0548VANDERBORGHT Bram FP7.6 P0897VASSILJEVA Kristina FP3.2 P0455VELAPPA Ganapathy TI.17 P0782VELTKAMP Remco C TM6.2 P0495VELUVOLU Kalyana C FI.4 P0775VENKATESH Svetha FP2.1 P0212VESAOJA Eero WM4.2 P0514VIJAY Avinash WM2.4 P0850VIJAY Vaidehi FI.12 P0418VIJAY Vaidehi FP6.6 P0472VIJAY Vaidehi TA7.6 P1127VIJAY Vaidehi TI.12 P0419VIJAY Vaidehi WI.32 P0405VIJAY Vaidehi WM7.6 P0615VIMAL M D TA7.6 P1127VINAGRE Blas M TI.5 P0669VINCKE Bastien WM7.5 P0889VIVET Damien WM7.3 P0785VOROBIEVA Helena TA2.6 P1054VRÁNA Stanislav TI.6 P1037

WWAHID Herman TA2.3 P0428WAN Li WI.10 P0558WAN MUDA Wan Mariam TP3.4 P0935WANG Bang WA4.3 P1135WANG Bin TP5.2 P1099WANG Chang-Qing FM1.2 P0258WANG Chi-Kai WI.36 P0179WANG Dan WP3.3 P0600WANG Danwei FA5.1 P1153WANG Danwei FI.17 P1158WANG Danwei TA5.1 P1151WANG Danwei TA6.6 P0976WANG Danwei TP4.3 P0962WANG Han FA5.1 P1153WANG Han TA6.3 P0315WANG Hee Lin WP7.4 P0991WANG Huawei FA3.2 P0313WANG Jack Jianguo WM7.4 P0883

146

WANG Jian-Gang FA6.6 P0586WANG Jian-Gang WP7.4 P0991WANG Jian-Gang WP7.5 P1148WANG Jingguo FI.16 P0940WANG Jun WI.1 P0309WANG Jun WI.10 P0558WANG Lei WM5.4 P0902WANG Liangshun WI.17 P0696WANG Liangshun WM3.4 P0500WANG Lichao WI.18 P0793WANG Liuping TP5.2 P1099WANG Qingchun FM5.5 P0759WANG Qiyuan FP7.5 P0724WANG Shyi-Wen FA4.5 P0664WANG Tao TA4.5 P0700WANG Tao TI.21 P0943WANG Weijiang FA7.2 P0769WANG Weiqun FM2.5 P0651WANG Weiqun WA3.3 P0727WANG Wenhui WI.20 P0621WANG Xiaofan WM3.5 P0529WANG Xiaolin WA7.5 P0445WANG Xiaoxin TA4.5 P0700WANG Xingcheng WP3.2 P0156WANG Xuan WI.5 P0767WANG Yong WM3.2 P0249WANG Youyi FP4.5 P0925WANG Yuechao FA5.5 P0352WANG Zhan TA3.5 P1010WANG Zuankai WA7.5 P0445WEBER Bernhard TP7.2 P0507WEE Yen Tat, Louis TP4.2 P0924WEI Guowu WA6.3 P1018WEN Changyun TA4.1 P0871WEN Changyun WP2.2 P0760WEN Yumei TA4.3 P0409WENG Yang WA4.4 P1144WEST Geoff A FM6.3 P0844WESTELL Jamie FM7.1 P0159WEYAND Christian FM4.1 P0243WIERING Marco A TM6.2 P0495WIJAYA Andika Aji TP5.1 P1089WIJESOMA W S FM5.1 P0842WIJESOMA W S TP6.5 P0770WIJESOMA W S WP6.1 P0300WILLIAMS Stefan WM8.3 P0661WONG Choon Yue WP5.3 P0589WONG Jia Yiing, Patricia FP5.4 P0764WONG Jia Yiing, Patricia FP5.5 P0765WONGWIRAT Olarn TA4.6 P0754WOO Seok WA3.2 P0706WU Chih-Hung WM5.5 P0484WU Haiyan TP7.6 P0913WU Jiayun FA2.3 P0987WU Panlong WI.2 P0340WU Qinghe FM4.2 P0470WU Quanjun WA2.1 P0332WU Xingming FA3.4 P0492WU Yan WM8.2 P0539

XXI Ning FA5.5 P0352XIANG Cheng FP5.2 P0554XIANG Lan FA2.1 P0371XIANG Lan WA2.1 P0332

XIAO Gaoxi TI.1 P0247XIAO Gaoxi WM2.1 P0197XIAO Nan WP4.2 P0434XIAO Wendong FP6.5 P1091XIAO Wendong WA4.4 P1144XIAO Yang WA3.2 P0706XIE Chengkang FM2.4 P0569XIE Jianping TP6.1 P0502XIE Lihua TA1.1 P0989XIE Lihua TA5.2 P1152XIE Lihua TP1.1 P0364XIE Lihua TP5.4 P1145XIE Lihua WA4.4 P1144XIE Lihua WI.12 P0835XIE Lihua WP4.2 P0434XIE Wenyu TI.33 P0580XIE Zongwu FM4.3 P0901XING Guolin WM2.3 P0552XING Yan FI.22 P0256XIONG Guangming FP2.3 P0623XIONG Guangming TM6.5 P0540XU De FA3.2 P0313XU Fen FP5.6 P0857XU Haoming FM3.2 P0535XU Huiling WA3.4 P1129XU Jiahe FM2.1 P0229XU Jian FA2.3 P0987XU Jian TA6.3 P0315XU Jianxin TP4.4 P0977XU Jun TA1.1 P0989XU Li WA3.1 P0689XU Lihong TA6.2 P0281XU Min WA8.1 P0224XU Ping FM7.4 P0665XU Wanying WI.37 P0372XU Wenfu TI.18 P0851XU Yuan FA1.5 P0278XU Zhao FI.17 P1158XUE Zhong TM2.5 P0755

YYAMASAKI Takumi FA2.5 P0248YAME Joseph J TA2.1 P0915YAME Joseph J TP3.6 P1058YAN Lei WM7.4 P0883YAN Yi FA5.1 P1153YANG Duanduan TM2.1 P0415YANG Guilin FA3.6 P0736YANG Guilin FI.25 P1104YANG Guilin WA5.1 P0656YANG Guilin WM2.6 P1105YANG Haibo TA6.6 P0976YANG Jin TA4.3 P0409YANG Peng FP6.1 P0894YANG Peng WM2.3 P0552YANG Ping FA3.2 P0313YANG Song TI.28 P0845YANG Xuebing WA5.2 P0744YANG Yi WI.18 P0793YANG Zhuo WP8.5 P1114YAO Juan FM2.5 P0651YAO Juan WA3.3 P0727YAO Juan WP3.4 P0966YAO Xiao-Lan FM4.2 P0470YAU Wei Yun FA6.6 P0586

147

YAU Wei Yun FI.31 P1102YAU Wei Yun WP7.4 P0991YAU Wei Yun WP7.5 P1148YE Shuxia FM2.5 P0651YE Shuxia WA3.3 P0727YE Yongqiang TA6.6 P0976YEH Ting-Fu WM5.5 P0484YEO Anthony WM6.5 P0908YI Lili WP2.3 P0867YING Ying TA6.3 P0315YOON Jongmin FM6.6 P1142YOU Keyou TP1.1 P0364YOU Yun WA4.1 P0985YOUSSEF Karim FP6.2 P0992YU Jiancheng FA5.4 P0523YU Jiancheng WI.27 P1073YU Ming TP4.3 P0962YU Naigong FP7.5 P0724YU Xinghuo TP5.2 P1099YU Xinguo FI.24 P1070YU Yongli WI.18 P0793YU Zhenyu FP7.4 P0662YUAN Changliang FP3.4 P0534YUE Haosong FA3.4 P0492

ZZAN Yao WP6.4 P0780ZARADER Jean-Luc FP6.2 P0992ZAVERI Mukesh A WM6.3 P0553ZENG Fanchao WM5.3 P0737ZENG Wenjun WI.10 P0558ZENGER Kai WM4.2 P0514ZHAI Chao WM3.1 P0210ZHAI Lainyin FP3.6 P0447ZHAI Xiaojun TA7.1 P1107ZHAN Ce FA6.4 P0605ZHANG Aiqun WI.27 P1073ZHANG Chengning WA8.1 P0224ZHANG Cishen WA3.4 P1129ZHANG Danhong TP4.3 P0962ZHANG Dawei WI.25 P0532ZHANG Gang FA2.1 P0371ZHANG Gang WA2.1 P0332ZHANG Hai-tao WM3.1 P0210ZHANG Hong TM2.6 P0847ZHANG Hongke FI.13 P0498ZHANG Huanshui WP4.5 P1069ZHANG Jie WI.37 P0372ZHANG Peng WA1.3 P1137ZHANG Rui Rui WA6.3 P1018ZHANG Sen FP6.5 P1091ZHANG Sen WA4.4 P1144ZHANG Wei WI.37 P0372

ZHANG Weizhong FI.15 P0125ZHANG Xiao-Dong FM4.2 P0470ZHANG Xiaolin WA2.4 P0597ZHANG Xiaozheng WI.35 P0872ZHANG Xueqiang FA5.4 P0523ZHANG Yan WM2.3 P0552ZHANG Yangyang FM2.4 P0569ZHANG Ye WI.13 P0874ZHANG Ying WI.37 P0372ZHANG Yu FM2.3 P0477ZHANG Yun FI.22 P0256ZHANG Zhengtao FA3.2 P0313ZHANG Zhifei WI.5 P0767ZHAO Chi FA7.2 P0769ZHAO Liang WM7.4 P0883ZHAO Liping TA6.5 P0587ZHAO Mingyang WA8.1 P0224ZHAO Shizheng FI.7 P1088ZHAO Tong TI.21 P0943ZHAO Weihua WM3.3 P0282ZHAO Weiwei FI.6 P0493ZHAO Xiaoguang WP5.4 P0602ZHAO Xiaoyu TI.25 P0609ZHAO Zhong WI.1 P0309ZHENG Lihong TA7.3 P1113ZHENG Tao FI.13 P0498ZHENG Yongjun FM4.3 P0901ZHENG Yu WP3.4 P0966ZHONG Jun FP4.3 P0471ZHONG Ling FI.22 P0256ZHONG Zhao Wei TP4.2 P0924ZHONG Zhaowei WM2.6 P1105ZHOU Changjiu FM5.1 P0842ZHOU Jin FA2.1 P0371ZHOU Jin WA2.1 P0332ZHOU Junhong TP4.1 P0923ZHOU Junhong TP4.2 P0924ZHOU Junhong TP4.4 P0977ZHOU Shengfeng WM5.6 P1026ZHOU Shou-Han FM5.2 P0981ZHOU Suiping TM2.2 P0619ZHOU Suiping WM5.3 P0737ZHOU Weixing WI.13 P0874ZHOU Yucheng FM2.1 P0229ZHOU Zhi WP7.3 P1147ZHU Changan FP4.3 P0471ZHU Jifeng FP2.2 P0431ZHU Qunxiong FA1.5 P0278ZHU Senqiang TA5.1 P1151ZINOBER Alan TA1.5 P0323ZOEBEL Dieter FM4.1 P0243ZONG Cong WM8.5 P0969ZU Linan FP6.1 P0894ZU Linan WM2.3 P0552

ICARCV 2010 PROGRAM OVERVIEW

Tuesday, 7 December 2010, 6 pm to 8 pm: Welcome ReceptionFoyer of Waterfront Ballroom, Level 2

Wednesday, 8 December 2010

0815 – 0845 Coffee-Tea with light refreshment, Foyer, Level 40845 – 0915 Opening Ceremony, Grand Ballroom, Level 40915 – 1015 Keynote Address 1: Audio-Visual Joint Processing for Active Object Detection, by Professor Zhengyou Zhang, Microsoft Research, USA (Grand Ballroom, Level 4)

Grand Ballroom Canary I Canary II Kingfisher Nightingale Pelican Oriole Bluebird Foyer1015 – 1040 Morning Tea & Interactive Session, Foyer, Level 4

WIInteractive Session

(C)309, 340, 414,671, 767, 982,

361, 1053, 185,558, 753, 835,874, 884, 960,

1080, 696, 793,378

Page 54

(R)891, 621, 526,567, 577, 896,

532, 748, 1073,692, 784, 1157,

394, 405Page 58

(V)151, 266, 872,179, 372, 399,

638, 907, 1063,667

Page 60

1040 – 1220

WA1 (IS)Advances in

Biometric Theory andApplications I

Dr Wei Yun YAU1143, 1124, 1137

1146Page 22

WA2 (C)Neuro-Fuzzy

Control IProf N. SundararajanProf Xiaolin ZHANG

332, 632, 141597, 979Page 23

WA3 (IS)Multidimensional

Systems Modeling andSignal Processing

Dr Yun ZOU689, 706, 727

1129, 425Page 24

WA4 (C)Sensor Network

SystemsProf Lihua XIEProf B-H SOONG

985, 994, 11351144, 513Page 25

WA5 (A)Man-machineInteractions

Prof J ROSSMANNDr Wenjie CHEN

656, 744, 826909, 800Page 27

WA6 (R)Modeling andIdentification

Prof Masoud SHAFIEEProf L D SENEVIRATNE

704, 10121018, 1060, 356Page 28

WA7 (R)Robot Control I,

Prof C C CHEAHProf Adel AKBARIMAJD

509, 584, 267717, 445Page 30

WA8 (V)Vision Applications

Prof Eric SungProf Ying SUN

224, 918944, 881, 349Page 31

1220 – 1330 Lunch, Waterfront Ballroom, Level 2

1330 – 1530

Panel SessionImaging and

Vision: Challengesand Opportunities

Prof Kar-Ann TOH

Page 16

WM2 (C)Process Control

Prof Vaidehi VIJAYProf Peng YANG

197, 223, 552850, 926, 1105Page 33

WM3 (C)Co-operative Control

Prof Kamesh SUBBARAProf Joseph J YAME

210, 249, 282500, 529, 1071Page 35

WM4 (C)Adaptive Control

Prof Koshy GEORGEProf Kai ZENGER

422, 514, 816904, 905, 876Page 36

WM5 (A)Process

Automation

Prof Ml DE LA SENProf Manish M PATIL

106, 245, 737902, 484, 1026Page 38

WM6 (R)Mobile Robotics I

Prof Maulin M JOSHI227, 357, 553795, 908, 984Page 39

WM7(R)Localization,Navigation,Mapping I

Prof Keck Voon LING320, 475, 785883, 889, 615Page 41

WM8 (R)Human Centered

Systems

Prof Jean-Yves FOURQUET691, 539, 661

965, 969, 1065Page 43

1530 – 1620 Afternoon Tea & Interactive Session, Foyer, Level 4

1620 – 1800

WP2 (C)Identification and

Estimation I

Prof K E HÄGGBLOMProf Zhiping LIN

616, 760, 8671035, 321Page 44

WP3 (C)Complex Systems

Prof Somanath MAJHIProf Yan-Ping LI

381, 156, 600966, 801Page 46

WP4 (IS)NetworkedSystems

Dr Ling SHI403, 434, 483

825, 1069Page 47

WP5 (R)Search, Rescue

and FieldRobotics

Prof Jose M GIRON-SIERRA

Dr Vijay Rajpurohit265, 423, 589

602, 951Page 48

WP6 (R)Mobile Robotics II

Prof Sai Hong TANG300, 1013, 347

780, 272Page 50

WP7 (IS)Advances in

Biometric Theoryand Applications II

Prof M TISTARELLIDr Wei Yun YAU

735, 1121, 1147991, 1148Page 51

WP8 (IS)Advanced Concept forEmerging Computer

Vision

Prof Siu Yeung CHO1110, 1106, 1108

1111, 1114Page 53

A: Automation; C: Control; IS: Invited Session; R: Robotics; V: Vision

ICARCV 2010 PROGRAM OVERVIEW

Thursday, 9 December 2010

0800 – 0830 Coffee-Tea with light refreshment, Foyer, Level 40830 – 0930 Keynote Address 2: Network based Control and Estimation Problems, by Professor Minyue Fu, University of Newcastle, Callaghan, Australia (Grand Ballroom, Level 4)

Grand Ballroom Canary Kingfisher Nightingale Pelican Oriole Bluebird Foyer0930 – 1000 Morning Tea & Interactive Session, Foyer, Level 4

TIInteractive Session

(C)247, 389, 453,

572, 669, 1037,709, 541, 565,

639, 1068, 419,Page 87

(R)517, 610, 682,690, 782, 851,413, 512, 943,

537Page 89

(V)306, 437, 609,790, 832, 845,

1028, 1130, 1001,407, 580Page 91

1000 – 1200

TA1 (C)Robust Control I

Prof Pierre APKARIANProf Hisaya FUJIOKA

989, 225, 603370, 323, 368Page 63

TA2 (C)Intelligent Systems

Dr Philippe MorignotDr Insu SONG

915, 427, 428505, 978, 1054Page 64

TA3 (R)Localization, Navigation

and Mapping II

Prof K SRIDHARANDr Jack Jianguo WANG

342, 402, 406677, 1010, 1103Page 66

TA4 (A)Instrumentation

Systems

Prof Changyun WEN871, 1049, 409596, 700, 754Page 67

TA5 (IS)Collaborative UAV and

UGV Systems

Prof Patrick DOHERTY1151,1152,1154

1155,1156Page 69

TA6 (V)Image/video Analysis I

Prof Mani Maran RATNAMProf Yongqiang YE

254, 281, 315346, 587, 976Page 70

TA7 (IS)Computer Vision &Pattern Recognition

Dr Soodamani RAMALINGAM1107, 1109, 1113

429, 548, 1127Page 72

1200 – 1310 Lunch, Waterfront Ballroom, Level 2

1310 – 1510

Panel SessionControl andAutonomy:

Challenges andOpportunities

Prof Yeng Chai SOH

Page 19

TM2 (V)Feature Extraction,

Grouping andSegmentation I

Prof Kezhi MaoDr Andrzej SLUZEK

415, 619, 713731 755, 847Page 74

TM6 (V)Computational

Intelligence in Vision

Prof Eric Sung351, 495, 10451056, 540, 895Page 75

1510 – 1600 Afternoon Tea & Interactive Session, Foyer, Level 4

1600 – 1800

TP1 (4-5.20pm) (IS)Sensor Networks

Prof Han-Lim ChoiDr Wendong Xiao

364, 799, 8111117

Page 77

TP2 (IS)Advances in

Biometric Theoryand Applications III

Prof Kar-Ann TohDr Wei Yun Yau

798, 1132, 1149719, 726Page 78

TP3 (C)Robust Control II

Prof S JanardhananProf Krzysztof J LATAWIEC

380, 386, 1020935, 673, 1058Page 79

TP4 (IS)Intelligent Diagnosis

and Prognosis

Ms Junhong ZHOU923, 924, 962

977, 1115, 1128Page 81

TP5 (IS)Variable StructureControl Systems

Dr Suiyang KHOO1089, 1099, 11181145, 1047, 263Page 82

TP6 (IS)Autonomy of Vehicular

Systems I

Prof. Christian LaugierDr. Philippe Martinet

502, 511, 655684, 770, 520Page 84

TP7 (R)Visual Servoing

Prof Wen-Chung CHANGDr Jorge POMARES

396, 507, 560611, 688, 913Page 85

1900 – 2200 Conference Banquet, Grand Ballroom, Level 4A: Automation; C: Control; IS: Invited Session; R: Robotics; V: Vision

ICARCV 2010 FINAL PROGRAM OVERVIEW

Friday, 10 December 2010

0800 – 0830 Coffee-Tea with light refreshment, Foyer, Level 40830 – 0930 Keynote Address 3: Robots Moving Closer to Humans, by Professor Bruno Siciliano, University of Naples Federico II, Italy (Grand Ballroom, Level 4)

Grand Ballroom Canary Kingfisher Nightingale Pelican Oriole Bluebird Foyer0930 – 1000 Morning Tea & Interactive Session, Foyer, Level 4

FIInteractive Session

(C)304, 385,

775, 986, 493,1088, 1095, 142,

333, 488, 418,498, 1067Page 125

(R)125, 940, 1158,324, 890, 1085,

961Page 127

(V)256, 1055, 1070,1104, 482, 601,648, 114, 882,

1102Page 129

1000 – 1200

FA1 (C)Nonlinear Systems

Prof Bill GOODWINEProf Maryam DEHGHANI

147, 1029, 1096231, 278, 757Page 93

FA2 (C)Networked Control

systems

Prof Young Soo SUHProf Gang ZHANG

371, 934, 9871023, 248, 467Page 95

FA3 (IS)Robotics in Unstructured

Environments

Prof XiaoQi CHENProf De XU

308, 313, 360492, 528, 736Page 96

FA4 (A)Intelligent Automation

Prof Manish M PATILProf Yongduan SONG

624, 714,828869, 664, 906Page 98

FA5 (IS)Autonomy of

Vehicular Systems II

Prof Philippe MartinetProf Han WANG

1153, 1120, 1150523, 352, 516Page 99

FA6 (V)Biometrics

Prof Eric SUNGDr Xiaozheng ZHANG

672, 937, 171605, 812, 586Page 101

FA7 (V)Image/video Analysis II

Prof Eam Khwang TEOHProf Yongqiang YE

490, 769, 879916, 188Page 102

1200 – 1310 Lunch, Waterfront Ballroom, Level 2

1310 – 1510

FM1 (C)Control Applications I

Prof Kirsten MoelgaardNIELSEN463, 258, 824

1024, 1032, 1064Page 104

FM2 (C)Delay Systems

Prof Pedro ALBERTOSProf Chengkang XIE

229, 438, 477569, 651, 1131Page 105

FM3 (R)Mobile Robotics III

Prof Sai Hong TANGProf Yongduan SONG

421, 535, 660680, 1038Page 107

FM4 (A)Factory Modeling and

Simulation

Prof Dieter ZOEBELDr Daoping Li

243, 470, 901917, 620, 831Page 108

FM5 (IS)Human Robot

Interaction

Dr Sarath KODAGODAMr Rajesh Elara MOHAN

842, 981, 9931042, 759Page 110

FM6 (V)Tracking andSurveillance

Prof Geoff A WEST702, 946, 844

852, 329, 1142Page 111

FM7 (V)Pattern Recognition

Prof N Sundararajan N159, 571, 592665, 817, 945Page 113

1510 – 1600 Afternoon Tea & Interactive Session, Foyer, Level 4

1600 – 1800

FP1 (4-5.20pm) (IS)Underwater Robotics

Dr Norimitsu SAKAGAMI959, 967, 1000

1036Page 114

FP2 (V)Feature Extraction,

Grouping andSegmentation II

Prof Tele TANDr Frederic MAIRE

212, 431, 623681, 1090, 659Page 115

FP3 (C)Neuro-Fuzzy Control II

Prof Meng Joo ERDr Xiang LI

983, 455, 1008534, 743, 447Page 117

FP4 (C)Control Applications II

Dr Arun D MAHINDRAKARDr Wen-Jong LIN

833, 146, 471604, 925, 955Page 119

FP5 (C)Identification and

Estimation II

Prof Patricia WONGProf Salim AHMED

687, 554, 614764, 765, 857Page 120

FP6 (R)Robot Sensing,

Perception and DataFusion

Prof Vaidehi VIJAYDr Sen ZHANG

894, 992, 1046958, 1091, 472Page 122

FP7 (R)Robot Control II

Prof Danwei WANG268, 515, 658662, 724, 897Page 123

A: Automation; C: Control; IS: Invited Session; R: Robotics; V: Vision