Visualisation and Generalisation of 3D City Models

Bo Mao

August 2010

TRITA SoM 2010-08

ISSN 1653-6126

ISRN KTH/SoM/-10/08/-SE

ISBN 978-91-7415-715-4

ii

© Bo Mao 2010

Licentiate Thesis

Geoinformatics Division

Department of Urban Planning and Environment

Royal Institute of Technology (KTH)

SE-100 44 STOCKHOLM, Sweden

iii

Abstract 3D city models have been widely used in different applications such as urban planning, traffic control, disaster management etc. Effective visualisation of 3D city models in various scales is one of the pivotal techniques to implement these applications. In this thesis, a framework is proposed to visualise the 3D city models both online and offline using City Geography Makeup Language (CityGML) and Extensible 3D (X3D) to represent and present the models. Then, generalisation methods are studied and tailored to create 3D city scenes in multi-scale dynamically. Finally, the quality of generalised 3D city models is evaluated by measuring the visual similarity from the original models. In the proposed visualisation framework, 3D city models are stored in CityGML format which supports both geometric and semantic information. These CityGML files are parsed to create 3D scenes and be visualised with existing 3D standard. Because the input and output in the framework are all standardised, it is possible to integrate city models from different sources and visualise them through the different viewers. Considering the complexity of the city objects, generalisation methods are studied to simplify the city models and increase the visualisation efficiency. In this thesis, the aggregation and typification methods are improved to simplify the 3D city models. Multiple representation data structures are required to store the generalisation information for dynamic visualisation. One of these is the CityTree, a novel structure to represent building group, which is tested for building aggregation. Meanwhile, Minimum Spanning Tree (MST) is employed to detect the linear building group structures in the city models and they are typified with different strategies. According to the experiments results, by using the CityTree, the generalised 3D city model creation time is reduced by more than 50%. Different generalisation strategies lead to different outcomes. It is important to evaluate the quality of the generalised models. In this thesis a new evaluation method is proposed: visual features of the 3D city models are represented by Attributed Relation Graph (ARG) and their similarity distances are calculated with Nested Earth Mover’s Distance (NEMD) algorithm. The calculation results and user survey show that the ARG and NEMD methods can reflect the visual similarity between generalised city models and the original ones. Keywords: 3D city models, Visualisation, CityGML, X3D, Generalisation, Aggregation, Typification, Quality evaluation

iv

Acknowledgements I am grateful to my supervisors, Professor Yifang Ban, KTH and Associate Professor Lars Harrie, Lund University, for their constant support, instruction, guidance, facilitation and encouragement during the research and daily life throughout the duration of my study in Sweden. Thanks all the staff and PhD students at Geoinformatics Division, KTH. The united international team gives me inspiration and feelings of home. I am also grateful to Helen Rost from Blom AB and Peter Axelsson from Digpro AB for the 3D data in the research. Erik Olsson and Hongchao Fan also assisted in this thesis research. I would like to thank China Scholarship Council and School of Architecture and Built Environment at KTH for providing the scholarship (2008-2012). I also thank the school of Computer Science and Engineering, Southeast University, China for the chance to study abroad. This Work has been carried out as part of a ViSuCity research project funded by the Knowledge Foundation, Swedish Foundation for Strategic Research, Vinnova, The Swedish Foundation for Health Care Sciences and Allergy Research, and Invest in Sweden. Finally, I would like to thank my family and my friends for their support all along. Bo Mao Stockholm, August 2010

v

Table of Contents 1. Introduction ....................................................................................................... 1

1.1 Background .................................................................................................. 1

1.1.1 Use of 3D city model............................................................................. 1

1.1.2 Related technologies of 3D city model ................................................. 2

1.1.3 The current research status of 3D city models...................................... 4

1.2 Research objectives...................................................................................... 7

1.3 Organisation of the thesis ............................................................................ 8

2. Literature review .............................................................................................11

2.1 Representation of 3D city models – CityGML..........................................11

2.1.1 CityGML overview .............................................................................11

2.1.2 CityGML module and levels of detail.................................................13

2.1.3 CityGML resources .............................................................................14

2.1.4 Other related standards........................................................................15

2.2 Visualisation of 3D city models.................................................................17

2.2.1 Introduction .........................................................................................17

2.2.2 X3D .....................................................................................................18

2.2.3 KML and other standards....................................................................19

2.2.4 Future development.............................................................................20

2.3 Generalisation of 3D city models ..............................................................22

2.3.1 Generalisation algorithm.....................................................................22

2.3.2 Generalisation evaluation....................................................................24 3 Related methodologies .....................................................................................25

3.1 Visualisation frameworks ..........................................................................25

3.1.1 Direct visualisation..............................................................................25

3.1.2 Visualisation using 3D standards ........................................................25

3.1.3 Visualisation of CityGML...................................................................28

3.2 Dynamic generalisation and visualisation .................................................31

3.2.1 Aggregation .........................................................................................31

3.2.2 Dynamic typification...........................................................................34

vi

3.2.3 Multiple representation data structure.................................................37

3.3 Generalisation evaluation ..........................................................................38

4 Results and Discussions ...................................................................................41

4.1 Introduction................................................................................................41

4.2 Visualisation framework............................................................................41

4.3 Multiple representation data structure .......................................................43

4.4 Generalisation results evaluation...............................................................45

4.5 Typification based on MST .......................................................................46

5 Conclusions and future research ......................................................................49

5.1 Conclusions................................................................................................49

5.2 Future research...........................................................................................50

References ...........................................................................................................51

vii

List of Figures

Figure 2.1 The five levels of detail (LODs) defined by CityGML

(source: IGG Uni Bonn) .....................................................................................14

Figure 2.2 Relationship among VDI 3805, IFC and CityGML .........................16

Figure 3.1 A topographic map (source: Eran et al., 2005).................................27

Figure 3.2 Workflow scheme of X3D model generation...................................27

Figure 3.3 The architecture of a web 3D service (source: Reitz et al., 2009)....28

Figure 3.4 The system architecture of the virtual 3D city model of Berlin (source: Döllner et al., 2006)...............................................................................29

Figure 3.6 Direct merge (Source: Bundy et al., 1995).......................................31

Figure 3.7 Snap merge (source: Bundy et al., 1995)..........................................32

Figure 3.8 2D building projections (source: Anders, 2005) ..............................33

Figure 3.9 Aggregation of 3D buildings (source: Anders, 2005) ......................33

Figure 3.10 Aggregation of ground plans ...........................................................34

Figure 3.11 Typification based on Kohonen Net (source: Sester, 2005)............35

Figure 3.12 Work flow chat of dynamic typification..........................................37

Figure 3.13 Example of P-tree (source: Coors, 2003)........................................38

Figure 4.1 Framework of the online 3D city system..........................................42

Figure 4.2 CityTree model .................................................................................43

Figure 4.3 Work flow chat of generalisation .....................................................44

Figure 4.4 Work flow chat of dynamic typification...........................................47

viii

List of Abbreviations 3DMLW - 3D Markup Language for Web ADE - Application Domain Extensions ARG - Attributed Relation Graph BIM - Building Information Modelling BLG-tree - Binary Line Generalisation tree CAD - Computer Aided Design CityGML - City Geography Makeup Language COLLADA- COLLAborative Design Activity DOM - Document Object Model EMD - Earth Mover’s Distance GAP-tree - Generalised Area Partitioning tree GML - Geography Markup Language GPU - Graphics Process Units HTML - Hypertext Markup Language IFC - Industry Foundation Classes ISO - International Organization for Standardization KML - Keyhole Markup Language LOD - Levels of detail MBR - Minimum Boundary Rectangle MST - Minimum Spanning Tree NEMD - Nested Earth Mover’s Distance O3D - Open 3D OGC - Open Geospatial Consortium U3D - Universal 3D VRML - Virtual Reality Modelling Language X3D - Extensible 3D XSLT - Extensible Stylesheet Language Transformations

ix

List of Papers This thesis is based on the following papers, which are referred to in the text by their Roman numbers. I. Bo Mao, Yifang Ban, Lars Harrie, 2009. A Framework of Online 3D City Visualization using CityGML and X3D, in the 6th International Symposium on Digital Earth (ISDE6), Beijing, China, September, 2009. II. Bo Mao, Yifang Ban, Lars Harrie, 2009. A Multiple Representation Data Structure for Dynamic Visualisation of Generalised 3D City Models, ISPRS Journal of Photogrammetry and Remote Sensing, special issue on Quality, Scale and Analysis Aspects of Urban City Models, accepted for publication III. Hongchao Fan, Bo Mao, Liqiu Meng, 2010. An automatic approach for the typification of façade structures, ISPRS Journal of Photogrammetry and Remote Sensing, special issue on Quality, Scale and Analysis Aspects of Urban City Models, revised version submitted IV. Bo Mao, Lars Harrie, Yifang Ban, 2010. A Dynamic Typification Method of 3D City Models using Minimum Spanning Tree, Proc. 6th international conference on Geographic Information Science (GIScience2010), Zurich, Switzerland, 14-17 September.

x

1

1. Introduction

1.1 Background

1.1.1 Use of 3D city model Along with the development of human society, an increasing amount of people are living in the cities, the home to half of the world’s population – about 3.3 billion. By 2030, the number will be near 5 billion (United Nations, 2008). Therefore, it is essential to develop efficient technologies to assistant the management of modern cities. The 3D city model related techniques are definitely required by number of communities of interest (Rich, 2009). 3D city models have already been widely used to assist urban management related applications such as urban planning, traffic control, mobile telecommunication designing, etc. Therefore, many countries and cities are creating and releasing their official 3D model. For example, a new 3D city model of Berlin has been available to the public since March 2009. It includes a total of 474,000 fully textured buildings spanning an area of 857 km2. This official Berlin model can be viewed through Google Earth and is available in CityGML format (Kada, 2009). The official Berlin 3D city model creates a basis to gather integrate and public existing geo-information, and a framework for specialist and non-specialist members of the public to take part in planning and decision-making processes. Landscape planners need 3D city models to visualise the impact of newly proposed projects on the city environment (Sadek et al., 2002). Murata (2004) developed a 3D-GIS application prototype to visualise the existing state of cities and to perform simulations of district development plan. In this application, the high-resolution 3D city model is combined with the 2D urban planning database for spatial analysis. The ViSuCity project (Ban et al., 2008) lead by Prof. Yifang Ban from KTH, Sweden is trying to develop an effective web-based, interactive 5D (3D + time + sustainability) visualisation demonstrator, ViSuCity, to support sustainable city planning in terms of information sharing, analysis, development, presentation and communication of ideas and proposals throughout the city planning processes. Security personnel need 3D city model to be able to conduct 3D line of sight analyses (Zhu et al., 1995), develop restricted area court on scenarios, etc. Ying et al. (2002) employs 3D city model to design the Closed-circuit television monitoring system. Public safety agencies such as fire department and emergency medical services need 3D city model among others to help them locate the available fire response equipment, manage the local transportation

2

infrastructure in the case of an emergency, and set the evacuated peoples, etc (Tang et al., 2006). Another example is Schulte and Coors (2008) that implemented a flood visualisation system based on 3D city models to illustrate the flood emergency. Space managers are interested in understanding the form, function, assignment, and availability characteristics of their space in 3D overtime. Dehghan and Steele (1997) talk about the necessity of using 3D city buildings to act as electromagnetic molds that define the size and shape of each three-dimensional cell to form a mobile radio network. Hermsmeyer et al. (2005) also use 3D city model to plan telecommunication antenna networks for highly developed inner-city areas of Kuwait. Commercial Real Estate community needs 3D city models to demonstrate their product not only at individual room level but also at the suite, building or surrounding area level. The researchers from University of Berlin use 3D city model to analysis and visualise the noise in North Rhine-Westphalia area (Czerwinski et al., 2007). Public administration agencies need 3D city models to regulate land use entitlement or administer local taxation policy in three dimensions. For example, the local taxes that are paid in Singapore are partly determined by the availability of sunlight in the occupied space and the impact of sunlight availability on the surrounding area (Rich, 2009). Also, the facilities, energy or public health management agencies will do a better job with the help of 3D city models. 3D city model is also wildly used in disaster control, location marketing, tourism etc (Schulze-Horsel, 2007). It is likely that 3D city models will play an important role in our daily life in the future.

1.1.2 Related technologies of 3D city model With the rapid development in computational capacities, computer graphics, web technologies and geospatial technologies, it is possible to create detailed 3D city models which can be simultaneously accessed by thousands of people from all over the world. First of all, computer devices become increasingly powerful. The famous Moore’s Law (Moore, 1965) indicates that the number of transistors that can be placed inexpensively on an integrated circuit has doubled approximately every two years. The trend has continued for more than half a century and is not expected to stop until 2015 or later (Kanellos, 2005). Nowadays not only the super computers but also PCs have the enough capacity of storing and processing huge volume of data. Modern Graphics Process Units (GPUs) are used to perform and accelerate calculations related to 3D computer graphics.

3

Mobile deceives, e.g. iphone, are also getting better and smarter to support 3D city models (Pairs3D, 2009). Second, computer graphics techniques are becoming more sophisticated. From the Hollywood movies to the PS3 video games, the virtual environments created by artists with modern computer graphics techniques are of such quality that even people can not identify them from real world. The process of creating 3D computer graphics can be sequentially divided into three basic phases: 1) 3D modelling which describes the process of forming the shape of an object; 2) layout and animation which describes the motion and placement of objects within a scene; 3) 3D rendering which produces an image of an object. 3D models especially the city or building models are now created by both man hands with tools e.g. CAD software and scanned into computer from real world objects. The 3D animation can be performed by motion capture system or simulated by physics engine. 3D rendering is usually performed by 3D computer graphics software or 3D Graphics API such as OpenGL (2010) or DirectX (2010). Third, the web techniques make it possible for anyone to access online 3D city model at anytime anywhere. According to the Nielsen's Law of Internet Bandwidth (Nielsen, 1998), a high-end user's connection speed grows by 50% per year. With the development of wireless network such as 3G/4G, Wi-Fi and IEEE 802.11, the Internet can be accessed from anywhere in quite high speed. The web service and cloud computing technologies supply access standards and computation power for the web 2.0 applications such as Facebook, Twitter or YouTube which bring huge number of users and create extremely amount of content. 25.6% of the world population, 1.7 billion, is using Internet by September 2009 (Internet World Stats, 2009) and 1 trillion unique URLs were found by Google at July 2008 (Google, 2008). The web is also going to the 3D. Currently, it is possible to display 3D content through common web browsers by installing certain plug-ins. The new makeup language HTML5, the next major revision of Hypertext Markup Language (HTML), will support video and 3D directly. That Apple’s new product iPad is not adopting Adobe Flash but HTML5 for multimedia shows the power and potential of HTML5. Meanwhile professional tools such as Google Earth, Virtual Earth and ArcGIS Explore also support online 3D city model. Fourth, geospatial technologies also make the online 3D city model possible. The remote sensing techniques like photogrammetry, high resolution aerial oblique photography, Lidar and GPS speed up the 3D city model creation process (Verma et al., 2006). All these progresses make the 3D city model available for everyone.

4

1.1.3 The current research status of 3D city models Along with the wide use of 3D city models and the development of related technologies, 3D city models have attracted increasing attention from both academic and industry fields. The current studies related to this thesis can be divided into three catalogues: model creation and management, visualisation and generalisation. 3D city model creation and management First of all, 3D city models have to be created before we can use them. Most of current 3D city models are created manually. Although there are a lot of tools to support the process, it still can not satisfy the needs for modelling large city area. Nowadays, an increasing amount of cities employ Lidar, digital camera and photogrammetry techniques to quickly generate textured 3D city models (Deng et al., 1995; Frueh, 2002; Frueh and Zakhor, 2003). 3D city models can also be converted from existing Building Information modelling (BIM) or CAD files (Döllnerand and Hagedorn, 2007). Since the 3D city models come from different sources, a standard is required to describe, store and exchange 3D city models among different parts, therefore CityGML (Kolbe et al., 2005; OGC, 2008) is designed. CityGML, issued by OGC in August 2008, is a common information model for the representation of 3D urban objects. It contains both geometric and semantic information. Since the huge mount of 3D city model data, databases such as Oracle and PostgreSQL are often be used to store and manage the data. Databases are supporting spatiotemporal data type and some of them, e.g. Oracle 11g, even support CityGML schemas. 3D city model visualisation Visualisation is a complex and important issue in 3D city environment. In most 3D city model related applications, visualisation is always a critical aspect, e.g. in urban planning and disaster mapping. In the context of the various applications, there are some of common visualisation requirements as follows. Visualisation of various levels of detail (LODs). It is often needed to visualise urban environment from different scales, e.g. from overview scale like a region down to detailed scale like a building or even a room. This kind of zooming function is required for interesting area selection, pattern indication, etc. Levels of detail (LODs) are proposed to represent models in different scales to reflect independent data collection processes with differing application requirements and facilitate efficient visualisation and data analysis.

5

Visualisation of changes over time. We need to see the status of the city model in different times to find out the changes. The temporal information of each object is stored to create a 4D (3D+time) city model which would be very useful in some applications such as urban planning. 3D city model generalisation In order to implement efficient visualisation, generalisation is required to reduce the data volume and improve the visual presentation. Generalisation stems from cartography and is the method whereby information is selected and represented on a map in a way that adapts to the scale of the display medium of the map, not necessarily preserving all intricate geographical or other cartographic details (Ruas, 2000). Traditionally, the cartographic generalisation is carried out manually. The experts from cartography create the generalised maps in different scales based their experience. However, with development of GIS related technologies such as satellite, GPS, and LiDAR, the volume of raw data especially for 3D city models is increased dramatically. These data are also required to be updated constantly or even in real-time. It is impossible to manually maintain real-time updated multi-scale representations for such amount of data even for a city not mention for a country. Therefore, the generalised methods discussed in this thesis are always carried out by computer instead of by humans. A number of researchers have paid attention to the generalisation of 3D city models. Different levels of detail (LODs) (OGC, 2008) are proposed to decrease the complexity of 3D models. Generalisation can be used to create 3D city model in lower LODs from higher LODs and visualise a large region of urban with reasonable amount of data which is very important in online 3D city model. Several algorithms have been proposed for generalisation of 3D city models for building, road etc. These algorithms not only reduce the data amount of the models, but also try to extract and enhance the interesting parts. In section 2.3.1, these existing algorithms are introduced. In section 3.3, some of the algorithms will be studied in detail. In chapter 4, they are compared with the generalisation methods proposed in paper II ~ IV. Since the generalised 3D city models are automatically created, it is essential to assessment the quality of these generalised models. The quality of the generalised 3D city model is depended on the quality of both original 3D models and the generalisation process (Harrie, 2001). Because the evaluation of generalisation quality mainly relies on the visual assessment of the results (Weibel and Dutton, 1999), this thesis is mainly concerned with the visual similarity between generalised models and the original ones. In Section 2.3.2, the current studies of generalisation quality evaluation is reviewed. In section

6

3.3, some evaluation methods are explained in detail. In section 4.4, these methods will be compared with the evaluation algorithm proposed in paper III. Other 3D city model related techniques Besides of the previous issues, there are several other 3D city model related techniques needed to be taken into consideration. Such as:

1. Analysis methods based on 3D city model. For example: line of sight analysis, noise modelling and 3D addressing/routing.

2. Mobile deployment. Many users would benefit from the ability to access a 3D city model in the field. This capability would be particularly useful to the security, public safety, facilities management, environmental quality, and municipal inspection communities.

3. Web publishing. Since the accessibility is very important to the 3D city model, we need to be able to consume these models as services published over the web. Given the complexity of some of 3D city models, server performance and network bandwidth constraints are significant challenges.

Focus of this thesis From the current status of online 3D city model, it is essential to develop efficient method to visualise 3D city models in various scales. Generalisation can be used not only to reduce the data volume and increase the visualisation speed but also automatic generate models in different LODs. This thesis is focusing on visualisation of multi-scale 3D city models using generalisation methods.

7

1.2 Research objectives The overall objective of this research is to develop a framework which can be used to visualise multi-scale 3D city models both online and offline efficiently. Specifically, the proposed framework enables importing city models in standard format e.g. CityGML from various sources such as file system, databases or even web services. It also generates suitable 3D city scenes of varying LODs according to the current user view point and recreates the scenes dynamically as user view point changes. The created scene should be interactively viewed both online and offline. The primary scientific questions are:

(1)How to represent and visualise 3D city model efficiently?

(2)How to automatically generate 3D city models in different LODs and implement dynamic visualisation of 3D city models in multi-scale?

(3)How to assess the visual quality of generalised 3D city models? The specific research objectives are:

(1)Make use of the 3D city model standard CityGML and create online and offline 3D scenes with X3D.

(2)Study current cartographic generalisation methods and propose new algorithms to support 3D city model generalisation.

(3)Propose suitable structure to represent the visual feature of 3D city models, and evaluate the similarity between generalised models and original ones.

This research is part of an academic PhD research project and has been carried out as part of the research project “ViSuCity - A Visual Sustainable City Planning Tool” (Ban et al., 2008). This project expects to contribute to a faster planning process with improved quality and enhanced sustainability and introduces interactive 3D visualisation as a de facto standard tool in planning and decision processes for municipal, regional and national projects.

8

1.3 Organisation of the thesis The thesis is structured as follows: Chapter 1 gives an overview of the research, including the background, research objectives, and organisation of the thesis. Chapter 2 reviews the state-of-the-art of research on 3D city model, CityGML, 3D visualisation standards and generalisation methods. Chapter 3 illustrates and compares with the existing methodologies including visualisation framework, generalisation algorithms and evaluation methods. Thereafter, Chapter 4 summarizes the main findings of the research. Chapter 5 concludes the research and explore future research directions. This thesis is based on the following papers, which are referred to in the text by their Roman numerals. I. Bo Mao, Yifang Ban, Lars Harrie, 2009. A Framework of Online 3D City Visualization using CityGML and X3D, in the 6th International Symposium on Digital Earth (ISDE6), Beijing, China, September, 2009. II. Bo Mao, Yifang Ban, Lars Harrie, 2009. A Multiple Representation Data Structure for Dynamic Visualisation of Generalised 3D City Models, ISPRS Journal of Photogrammetry and Remote Sensing, special issue on Quality, Scale and Analysis Aspects of Urban City Models, accepted for publish III. Hongchao Fan, Bo Mao, Liqiu Meng, 2010. An automatic approach for the typification of façade structures, ISPRS Journal of Photogrammetry and Remote Sensing, special issue on Quality, Scale and Analysis Aspects of Urban City Models, revised version submitted IV. Bo Mao, Lars Harrie, Yifang Ban, 2010. A Dynamic Typification Method of 3D City Models using Minimum Spanning Tree, Proc. 6th international conference on Geographic Information Science (GIScience2010), Zurich, Switzerland, 14-17 September. In paper I, the authors develop the framework of visualisation 3D city model. In this paper, CityGML is studied for 3D city models representation (Objective 1) and the CityGML file is converted to X3D scene which is viewed both offline and online (Objective 2). In paper II the authors propose a multiple representation data structure CityTree to represent building group and implemented the dynamic building aggregation (Objective 2 and 3). In paper III, a method for evaluation of generalisation results is proposed which can be used to calculate the visual similarity between the original facade structure and the generalised ones. This method is evaluated by user survey and the results shows

9

that it conforms to the human visual perception. This method can also be used for quality assessment of 3D city model generalisation (Objective 3). In paper IV, the authors typify the 3D city model using minimum spanning tree (Objective 2 and 3). The author of this licentiate thesis has made the implementation of Paper I, II, IV and evaluation method in paper III. Co-authors have given suggestions of methodologies, have assisted in writing and have implemented the building facade generalisation.

10

11

2. Literature review Along with the increasing demand for 3D city models, visualisation, as one of its most important research aspect, has drawn broad attentions from not only academic field but also industry authorities and public. Representation, presentation and generalisation of 3D models are the keys to effective visualisation of multi-scale 3D city models, which is discussed thoroughly in this chapter.

2.1 Representation of 3D city models – CityGML In order to store, manage and integrate 3D city models from multi-source, a standard is required to represent 3D city models. CityGML issued by OGC in August 2008 is the first international standard to represent 3D city models.

2.1.1 CityGML overview CityGML is a common information model for the representation of 3D urban objects. It is realized as an open data model and XML-based format for the storage and exchange of virtual 3D city models. As a new OGC standard, CityGML plays a leading role in the modularisation of urban geospatial information. CityGML is an application scheme for the Geography Markup Language version 3.1.1 (GML3) and represents both graphical appearance of 3D city models and their semantic properties. The current version of CityGML is 1.0 (OGC, 2008). Some applications have developed based on CityGML. However, CityGML is still evolving, new attributes and contents will be added into CityGML standard. CityGML Applications Several cities choose CityGML as their official 3D city models such as Stuttgart, Bonn and Berlin. They use CityGML to integrate existing 3D models of departments of business development and town planning to support urban planning, city marketing, and recruiting of new enterprises (Döllner et al., 2006). Meanwhile, CityGML has been used in environmental detection and protection. EU environmental noise mapping in North Rhine-Westphalia project is one of those (Czerwinski et al., 2006a; 2006b and 2007). The project makes use of CityGML to analysis the noise of the region. Also CityGML is used in

12

Homeland security and other geo-related analysis project such as Traffic Simulation and Driving Simulation with Municipal CityGML Geodata from Dusseldorf (Pantzer, 2008). CityGML Application Domain Extensions Since the city itself is quite complex, it is impossible for a standard to specify every detail of the city. Therefore, objects, which are not explicitly modeled in current version of CityGML, can be represented using the concept of generic objects and attributes. In addition, extensions to the CityGML data model applying to specific application fields can be realized using the Application Domain Extensions (ADE). Generics objects/attributes – allow extensions during runtime, but may cause arbitrarily and name conflicts because different self defined object may have the same name. The ADE can extend existing CityGML feature types; define new feature types and create its own XML schema definition. Therefore in most cases, ADE is chosen to extend CityGML. Some examples are:

CityGML noise ADE used in noise mapping project in North Rhine-Westphalia (Czerwinski et al., 2007). CityGML flood ADE for a web based 3D flood information (Schulte and Coors, 2008).

CityGML problems Since CityGML is a new standard under development, it also has some problems, which should be taken care of in the future research (Kolbe, 2007). 1. Size. CityGML files become very large (Several GB for the big city).

However, file sizes can be effectively reduced by gzip compression (reduced to around 10% of original size), but XML validation and processing can be problematic (classical DOM parsing not feasible due to main memory limitations) and WFS access might have to be realized in an asynchronous way in order to avoid timeouts.

2. Complex data model. Because the city itself is complex, the CityGML is complex. The CityGML specification is growing. This complexity makes it hard to implement all specifications.

13

2.1.2 CityGML module and levels of detail Since CityGML covers broad thematic fields of city object from geometrical, topological to semantic aspects, applications are not required to support all thematic fields of CityGML, but only a subset. The use of logical subsets of CityGML reduces the complexity of 3D city models. In CityGML standard version 1.0, possible subsets are defined and embraced as CityGML modules. CityGML consists of a core module and thematic extension modules. CityGML core module defines the basic concepts and components of the CityGML data model. The core module as the lower bound of overall CityGML data model and a dependency of all thematic extension modules must be implemented by any conformant system. Based on the core module, eleven thematic extension modules are introduced by version 1.0 of the CityGML standard. These modules are: Appearance, Building, CityFurniture, CityObjectGroup, GenericCityObject, LandUse, Relief, Transportation, Vegetation, WaterBody, TexturedSurface. These extension modules can be arbitrarily combined according to the requirements of the application. A combination of modules is called a profile. The union of all defined modules is CityGML base profile which forms the upper bound of the overall CityGML data model. The CityGML modules and base profile will be extended in the future by expert in the correspond fields. The application data can also be incorporated within the existing modules by using the concepts of generic city objects and attributes or with CityGML’s Application Domain Extension mechanism. In geometric aspect, CityGML supports 5 Levels of Detail (LODs) to reflect independent data collection processes with differing application requirements and facilitate efficient visualisation and data analysis (Figure 2.1). LOD0 is a two and a half dimensional Digital Terrain Model. LOD1 is the blocks model comprising prismatic buildings with flat roofs. LOD2 contains differentiated roof structures and surfaces. LOD3 is the architectural models with detailed wall and roof structures. LOD4 combines a LOD3 model with interior structures. City objects can be represented in all LODs simultaneously.

14

LOD0 LOD1

LOD2 LOD3 LOD4 Figure 2. 1: The five levels of detail (LODs) defined by CityGML

(source: IGG Uni Bonn)

2.1.3 CityGML resources As an open standard, CityGML is receiving wide support from academic fields and industry. Many software both open source and commercial are developed for CityGML visualisation, manipulation and management. For visualisation, Aristoteles (2009), an open source CityGML viewer, is developed by the Institute for Cartography and Geoinformation, University of Bonn. LandXPlorer CityGML Viewer 2009a (2009) is developed by the company Autodesk. FZKViewer 0.9 (2009) developed by the Institute for Applied Computer Science, KIT – University of the State of Baden-Württemberg and National Laboratory of the Helmholtz Association can support CityGML 0.4.0, CityGML 1.0.0, the CityGML Noise ADE, the CityGML Subsurface Structure ADE and Industry Foundation Classes (IFC) models. Other

15

software such as CityVu (2009), Ptolemy3D (2010), BS Contact Geo 7.2 (2009), FME 2010 Data Inspector (2009) also support CityGML visualisation. For manipulation, researchers from University of Applied Sciences Gelsenkirchen developed a Google Sketchup plug-in to import and export CityGML for editing. CityGMLl4j (2009), an open source Java class library and API for facilitating work with the CityGML, is developed by Institute for Geodesy and Geoinformation Science of the Berlin University of Technology. It can be used to read, process, and write CityGML datasets, and to develop CityGML-aware software applications. For management, 3DCityDBv2 (2009), a free and Open Source 3D geo database to store, represent, and manage virtual 3D city models on top of the Oracle 10G R2 / 11G spatial relational database, is developed by Institute for Geodesy and Geoinformation Science of the Berlin University of Technology. The database model contains semantically rich, hierarchically structured, multi-scale urban objects facilitating complex GIS modeling and analysis tasks. This thesis makes use of CityGML4j to implement the reading, parsing and exporting of CityGML files.

2.1.4 Other related standards Besides CityGML, there are some other international standards related to 3D city models such as VDI 3805 (2010) and Industrial Foundation Class (IFC, 2009). VDI 3805 is an industry standard for product data exchange in Building Services and is used as uniform format to model all Building Services equipment (heating technology, ventilation and air conditioning technology, sanitary engineering, etc). IFC is a neutral and open specification to facilitate interoperability in the building industry and is a commonly used format for Building Information Modelling (Eastman, 2009). IFC is now an official International Standard ISO/IS 16739 and has been made as compulsory format for publicly aided building projects by some countries e.g. Denmark. The relationship among VDI 3805, IFC and CityGML is shown in Fig. VDI 3805 is a standard about components in the building; IFC is focus on building modelling and CityGML is designed in the city level and contains more city objects such as vegetation, road etc.

16

Figure 2. 2: Relationship among VDI 3805, IFC and CityGML

These standards can be converted into each other. With some software e.g. ETU VDI 3805 Navigator (2008), models in VDI 3805 format can be saved as IFC format. IFC models can also be converted to CityGML models. The Open Source BIM Server project (BIMServer, 2010) now supports the export of CityGML. Models are converted on the fly from the stored and managed IFC models and are exported as CityGML LOD4 building models preserving their semantic information (thematic structure and attributes). A complete IFC model can also be exported into CityGML or contrary by using the IfcExplorer (2007). Since VDI 3805 and IFC models are designed for more detailed level, CityGML is the most suitable standard to represent 3D city models. Meanwhile, the data in these related formats can be converted into CityGML data and becomes an important source for 3D city models creation.

17

2.2 Visualisation of 3D city models In previous section, we described representation of 3D city models and especially the CityGML format. This section concentrates on visualisation (presentation) of 3D city models. CityGML is designed to represent 3D city models, but not to present or visualise 3D city models directly. As Kolbe (2008, p28) pointed out: “CityGML is complementary to visualisation standards like X3D or KML. While these address presentation, behaviour, and interaction of 3D models, CityGML is focused on the exchange of underlying urban information behind 3D objects”. Therefore, specific 3D visualisation techniques is required more efficiently 3D city models presentation.

2.2.1 Introduction In the early stage of computer development, visualisation was implemented directly based on the hardware (Carlson, 2003). The users had to solve everything. For example, if one wants to draw a line on the screen, one should not only give the coordination of start and end point but also consider the monitor resolution, frame buffer, and related system calls. Since the computer platforms were quite different from each other, the portability of the visualisation program at this stage was quite low. In order to simplify the development and increase the portability, standard APIs for writing applications that produce 2D and 3D computer graphics were proposed (DirectX, 2010; OpenGL, 2010). These APIs deal with the hardware, supply the function calls which can be used to draw complex 3D scenes from simple primitives. The manufactures can also upgrade their hardware without influencing the applications running on them by supporting the APIs. However, there were still problems. First, the APIs were still too complicated (OpenGL API contains 250 different function calls) and too primitive (only support atomic geometric objects e.g. point and linesegment) to implement 3D visualisation directly. In order to create 3D scenes, the user still was required to calculate the projections, transformations, render the scene and so on, which was not suitable for the complex 3D models like 3D city models. Second, there were several graphic APIs such as OpenGL, DirectX, Mesa 3D (2010), VirturalGL (2010), RISpec (2010), Glide (2010) and so on. Different versions have to be created to support different APIs. For 3D city models which needed be viewed by anyone from anywhere, it is not a wise choice to employ these basic graphic APIs for online 3D city models visualisation.

18

Along with the development of Internet, the requirement for sharing information is increasing. In order to represent the 3D computer graphics which can be viewed everywhere through Internet, standards are proposed. These standards describe the 3D scenes, and supply APIs for user interactivity. The visualisation of 3D city models can be saved in these standards which are supported by various platforms. Currently, there are several 3D visualisation standards such as VRML (Bell et al., 1995), X3D (2010), 3DMLW (2010), COLLADA (2010), KML (2010), O3D (2010) and U3D (2010).

2.2.2 X3D X3D is open standard file format and run-time architecture to represent and communicate 3D scenes and objects based on XML. It is the successor of VRML to encode the 3D scene using XML syntax as well as the Open Inventor-like syntax of VRML97, or binary formatting, and provide an enhanced APIs. X3D is proposed and managed by Web3D Consortium, and is now an ISO ratified standard that provides a system for the storage, retrieval and playback of real time graphics content embedded in applications, all within an open architecture to support a wide array of domains and user scenarios. X3D is particular designed for visualisation 3D scenes through Internet. X3D is now being used in many cases, such as e-learning (Thomas, 2008), medication (Jung et al., 2008; Willis, 2007; Hamza-Lup, 2006) and geo-related applications. Araujo et al. (2008) proposed a framework for 3D web-based visualisation of HLA-compliant (High Level Architecture) Simulations. X3D is also used for virtual cultural heritage (Eliens et al., 2007; Cabral et al., 2007) and Building Information Modeling (Campbell, 2007). In mobile GIS, 3D maps and models are implemented in X3D (Nurminen, 2006; 2007; Alessandro et al., 2007). Meanwhile, X3D itself is improving. Jung and Behr (2008) proposed a method to extend the animation; Buttussi et al. (2007) and Patterson (2007) focused on the search of X3D models. Bilasco (2007) and Pittarello (2006) added semantic information into X3D; Jung (2007) used X3D into Mixed Reality appliances; Weber and Parisi (2007) designed an open protocol for Wide-area Multi-user X3D. As an open standard, there are several tools to support the application of X3D. Many viewer software and plugins for blowers are free available for X3D visualisation both locally and online, for example Bit Management Contact viewer (2010), Octaga (2010), Flux (2007) and InstantReality (2010). Editors are also developed to create X3D file. X3D-Editor (2010) is a graphics file

19

editor for X3D that enables simple error-free editing, authoring and validation of X3D or VRML scene-graph files. X3D-Editor (2010) is the officially editor recommended by Web3D. Other editors such as Vivaty Studio (2010), BS Editor (2010), SwirlX3D Editor (2010), Flux Studio (2007) etc. are commercial editors. Some of the 3D software like Blender, 3DSMax, AutoCAD and Maya are also support any X3D file import/export. For development, Xj3D (2009) is an open source toolkit for X3D related applications to create interactive 3D scenes. Xj3D supports two types of visualisation platform Java Application and Java Applet which could be used to deliver 3D city models through Internet. ChefX3D (2007) focus on developing a toolkit for building 3D editors. It’s expected these editors will form part of a larger application. Typically these editors will be customized for authoring certain types of content. Rez (2007) is an open source framework and tools for translating gridded data, mainly geospatial, to different formats including images and multi-resolution models for X3D or VRML web browsing. Jeospace (2010) is another open source java toolkit for creating geo-spatial simulations. X3D is playing an important role in geo-visualisation applications. In Web3D Consortium, some researchers develop an X3D profile for GeoVisualisation, which is called X3D Earth (2010). X3D Earth is an open standards-based infrastructure for visualising information in a geospatial context. For example, the new tags related with geography are defined such as GeoOrgin, GeoCoordinate and GeoLocation. X3D Earth profile has been used in many projects and applications e.g. Terrain Generated by Rez, JeoSpace globe generator, NASA WorldWind (2008) with an X3D loader, Digital Nautical Chart (2010), Planet 9 Virtual Cities (2010), MBARI Monterey Bay operations (MBARI, 2010), NPS Savage Studio scenario creation (X3D Earth, 2010). Meanwhile, Web3D Consortium and OGC have made an agreement to build cohesive standards for Location-Based 3D web visualisation services, which contains the strengths of X3D as a 3D presentation/portrayal systems for OGC standards and how to integrate X3D with other OGC standards such as CityGML (Havele, 2010).

2.2.3 KML and other standards KML is a file format used to display geographic data in an Earth browser such as Google Earth and Google Maps. OGC has approved KML as an open standard in April 2008 and defined KML as an XML language focused on geographic visualisation, including annotation of maps and images. Although KML is not designed for 3D visualisation, it supports 3D model by COLLADA file. COLLADA is a COLLAborative Design Activity for establishing an

20

interchange file format for interactive 3D applications. COLLADA was created by Sony Computer Entertainment, is now an open standard XML scheme for exchanging digital assets among various graphics software applications. Therefore, it is mainly used in industry fields especially in game industry. Since KML/COLLADA is designed for Earth browser, X3D is a better choice to present online 3D city models for its compatibility with the HTML web and wide support from standard browsers such as Firefox or Internet Explorer. However, as the population of Google Earth and Google Map are increasingly popular, an increasing number of people not only view 3D city models from Google Earth but also create and share 3D city models through it. City of Berlin publishes its 3D city models through Google Earth in KML format (Kada, 2009). Many online GIS applications employ KML for visualisation. Shinozaki (2008) uses KML for virtual 3D urban design in Tokyo and Fukuoka. Calado et al. (2008) export hydrographic data product into KML. Chen et al. (2008) visualise the earthquake emergency with Google Earth. KML could be an important option for 3D city models visualisation and source for 3D city models generation. It is necessary to integrate KML with CityGML and X3D for better 3D city models visualisation. Besides X3D and KML/COLLADA, there are some other 3D visualisation standards e.g. 3DMLW, O3D and U3D. 3DMLW, stands for 3D Markup Language for Web, is an XML-based file format for representing 3D and 2D interactive content on the World Wide Web. Displaying 3DMLW requires 3DMLW plug-in to be installed on computer and it uses OpenGL for rendering. 3DMLW plug-in is developed by 3D Technologies R&D for common web browsers (Internet Explorer, Mozilla Firefox, Opera, etc.). O3D is an open source (BSD license) JavaScript API created by Google for creating interactive 3D graphics applications that run in a web browser window. Universal 3D (U3D) is a compressed file format standard for 3D computer graphics data. 3D objects in U3D format can be inserted into PDF documents and interactively visualised by Acrobat Reader since version 7. Compared with X3D and KML/COLLADA, these standards are designed for certain applications by different companies, but are not international standards approved by OGC or ISO and not widely supported by different communities. However, these standards can be used for certain purposes such as visualisation 3D city models in U3D to create 3D PDF files.

2.2.4 Future development

21

Currently, most of the 3D scenes in the web are rendered by plugins such as Adobe Flash, Microsoft Silverlight (2010), Java3D/JOGL Applet, O3D, X3D, KML/COLLADA and MPEG-4. These plugin-based systems have two major drawbacks. First, the user has to deal with plugin installation, security and browser or OS incompatibility issues, if the plugins are installed by default on the system. Second, the developers have to familiar with the details of different plugins to develop their applications. Users have to use plugins for 3D visualisation in the Web, because the Web itself does not support 3D right now. However, driven by the user needs, the new coming web standard HTML5 will improve the situation. HTML5 is the proposed successor of HTML 4.01, XHTML 1.0 and DOM Level 2 HTML. It aims to reduce the need for proprietary plug-in-based rich internet application technologies by supporting much data format including multimedia, 3D models etc. It is why Behr et al. (2009) proposed X3DOM that allows for integrating X3D nodes into HTML5 DOM content directly. This model is a similar approach to the current SVG integration model for 2D graphics. HTML5 and X3D create a promising future for online 3D visualisation. Therefore, X3D is a suitable choice to present online 3D city models.

22

2.3 Generalisation of 3D city models Visualisation of 3D city models in multi-scales is essential for online presentation. Zooming user interface is used to change the scale of viewed area in order to see more detail or less, and browse through different areas (Bederson and Hollan, 1994). It is important to find methods to generate lower LODs from higher LODs e.g. to generalise building groups in block level or city level not only to reduce the data volume but also hide the unnecessary details in different scales; these different representation could then be stored at different LOD in e.g. CityGML. In this section, some generalisation algorithms in 2D and 3D are first described. Then algorithms for measurement of the difference between the original models and the generalised ones are introduced. Finally, current multi-LOD fusion methods are presented to create continuous LOD.

2.3.1 Generalisation algorithm Automatic cartographic generalisation has been acknowledged as an important research area for a long time (Mackaness et al., 2007). An overview of generalisation methods is given in Harrie (2001). Generalisation contains several operators such as simplification, smoothing, aggregation, amalgamation, collapse, selection, typification, exaggeration, enhancement, classification, displacement (Shea and McMaster, 1989). Recently, much attention is paid on 3D city models related generalisation. In this thesis, the simplification and typification are mainly discussed since they are key aspects for generalisation buildings in 3D city models. The problem of simplifying 3D city models in higher LODs has been acknowledged for a long time (e.g. Köninger and Bartel, 1998). Much study has been performed on simplifying single buildings (Thiemann, 2002; Forberg, 2007; Kada, 2002; Sester, 2000; Fan et al., 2009). CityGML standard also specifies five LODs to satisfy the needs of efficient multi-scale visualisation. Several simplification methodologies have been proposed to obtain lower LODs required for efficient visualisation from higher LODs automatically. Meng and Forberg (2007) summarized generalisation methods to change the representation between different LOD. For the single building model simplification, Mayer (2005) and Forberg (2007) createed a scale-space technique partly based on the morphological operators opening and closing to simplify 3D building model. Half space model was used by Kada (2006) to detect the main outline of a building. Then he extends the simplification method to the roof structures using pre-defined roof types (Kada, 2007). As CityGML becomes the OGC specification, some researchers focus on simplify 3D city models stored in CityGML format. Fan et al. (2009) propose a method for efficient handling of

23

3D buildings modelled in CityGML LOD3 (corresponding to a detailed architecture model) and converted LOD3 model into LOD2 by using the exterior shell of the building model that drastically decreases the required number of polygons. They notice the huge gap between LOD2 and LOD3, and try to create LOD2.5 an intermediate level for more efficient visualisation. For group of buildings, aggregation is used to simplify the models. Aggregation means to combine individual buildings of city model in order to remove intermediate nodes or faces. The result of the aggregation is a reduced number of building objects and building surfaces. Anders (2007) introduces an aggregation method using 2D projections of linear building groups. Glander and Döllner (2007) make use of cell-based clustering to merge building blocks. Götzelmann et al. (2009) discuss the terrain-dependent aggregation of 3D city models. Typification is another effective method for 3D city models generalisation. Typification is the process of replacing a number of objects by a smaller number of new objects. It has been studied in 2D by e.g. Regnauld (1996), Moritz et al. (2009) and Anders and Sester (2000). Typification methods can be categorized into two types (Anders, 2005): with and without structural knowledge. Typification methods with structure knowledge try to detect the geometrical structures in the object groups and generalise the groups based on these structures. Regnauld (1996) employed minimum spanning tree (MST) to detect the linear structure of building groups and implements the typification generalisation based on the MST. Christophe and Ruas 2002 proposed another approach to detect the building linear structure to directly group the objects. Besides line structure, Burghard and Cecconi (2007) made use of Delaunay triangulation to detect the structures. Two dimensional structures or grid structures is also used for typification (Anders and Sester, 2000; Anders 2005). Fan and Meng (2010) make use of vertical and horizontal property of the windows to typify the facade. Most of these structure knowledge typification methods use Töpfer’s radical law (Töpfer and Pillewizer, 1966) to determine the reduction factor. Töpfer’s radical law provides numeric guidelines by which to determine how much detail to retain during map compilation and reduction. Typification methods without structural knowledge try to implement some common rules in the generalisation of the object groups. Müller and Wang (1992) used mathematical morphology to typify natural areal objects. They remove the smaller objects unless they are important. Sester and Brenner (2000) used a neural network—Kohonen Feature Maps to preserve the original structure. They move the remaining objects along the direction of removed ones to minimize a certain error measure. Then Sester and Brenner (2004) replaced every four nearby buildings with a new polygon connecting the centroids of the four removed buildings. But It may destroy the dominant structures of the models by using these methods (Anders, 2005).

24

2.3.2 Generalisation evaluation Generalisation of 3D city models could be implemented with a multitude of alternatives and produced different results. Even for the same generalisation operator, different strategies will lead to different outcome. Therefore, it is essential to evaluate these outcomes and select the most suitable as the generalisation result. In visualisation applications, the most important criterion is the visual similarity between the original 3D city models and the generalised ones. Generalisation evaluation methods have been studied by cartographic society for a long time. Shea and McMaster (1989) tried to find out which generalisation operator is more suitable at certain time by situation assessment of the conditions for generalisation, the measures indicating the need for generalisation and the controls on how to apply generalisation functionality. Harrie (2001) divided the generalisation quality evaluation into visual assessment, functional assessment and quantitative assessment; visualises the generalisation quality in a visual warning system. Zhang et al. (2009) quantitatively analyzed the map information of different scales in the range of the mapping scale by making use of landscape pattern index, and constructed an evaluation index system by the standard of cartographic generalisation evaluation. Filippovska et al. (2009) introduced a method to evaluate the quality of generalised building ground plans. Zhang et al. (2008) proposed an object-oriented density measurement based on skeletonisation of gap space to evaluate the spatial distribution density in map generalisation. Unfortunately, most of these generalisation quality evaluation methods are proposed for 2D cartographic maps. There are few studies on quality assessment of 3D city models generalisation. Since visual similarity is used as the main criterion to evaluate the generalisation results, techniques in pattern recognition can be used to measure the visual concept distance between the original 3D city models and the generalised ones. Mathematically, features of 3D city models can be represented as attributed relational graph (ARG), whose nodes contain the feature of objects in 3D city models e.g. buildings, and edges contain the relations between nodes e.g. distance between buildings. Therefore, the problem of quality assessment of generalisation can be converted to the matching of two ARGs which represent the original 3D city models and the generalised ones respectively. A great deal of effort has been devoted to develop efficient ARG matching algorithm. Kim et al. (2004) proposed a method using the nested structure of earth mover’s distance to calculate the difference between ARGs. The difference can be used to represent the visual similarity between 3D city models and evaluate the generalisation results.

25

3 Related methodologies

3.1 Visualisation frameworks Several frameworks have been proposed for visualisation of 3D city models. In this section, some of the typical frameworks are introduced in detail.

3.1.1 Direct visualisation Beck (2003) directly used OpenGL to visualise 3D city models in real-time. In order to get the high speed, this framework makes use of the computer graphics related technologies such as pre-stripping the triangles, frame controls, texture memory management etc. The data comes from database in self-defined 3D format and integrated with the textures. Several levels of detail are used to display different geometric representations of the same object at different times. Less detailed represented when the object is far away and more detailed when it is closer to the observer. A few web plugins for Internet Explorer are developed to facilitate the interactive visualisation of high-resolution 3D terrain and building data over the Internet. These plugins have to deal with TCP/IP protocols and be optimized for low-bandwidth data transfer. For security purposes all data streams are compressed and heavily encrypted. Although it may increase the visualisation speed, this direct visualisation framework has to deal with the basic computer graphics issues such as frame control, cloud rendering etc., which may increase development difficulty and workload dramatically. Different versions of the visualisation programs need be developed for different platforms. Too many details related with the basic operation system and graphics interfaces may highly influence the portability of the visualisation framework.

3.1.2 Visualisation using 3D standards Several studies have used 3D visualisation standards such as VRML, X3D and KML to present their 3D city models. Lerma and Garcia (2004), for example, employed VRML to visualise their historical centres 3D models. They developed the software TopVRML to model and join 3D virtual objects. TopVRML was written in C++ and made VRML code that can be used directly and interactively over the Internet or local networks. This software allows users to set the observer characteristics (flying, walking, or examine), sky and ground backgrounds; insert a 3D prototype and link it with web page; generate 3D

26

objects from images and create geometrical shapes such as cube, cylinder, cone etc.; and set the illumination by placing both directional and punctual colour lights in the scene. As the widely use of Google Earth and its visualisation language KML becoming an open standard of OGC, Google Earth has become a popular data viewer in a variety of fields including geoscience. Yamagishi et al. (2010) propose a 3D visualisation system for geoscience data based on KML. They developed a software package, called “KML generator” to convert original data files to KML files. These generators can deal with seismic tomographic models, geochemical data of rocks, and geomagnetic field models. With these KML generators, an overlapping visual presentation of different types of data can be obtained. Existing KML files of geoscience data (e.g., earth quake hypocenters and seismograph stations) provided by various research institutes around the world (such as the U.S. Geological Survey) can be also overlaid on these data. Eran et al. (2005) gave an example of creating and visualising 3D city models using X3D. Their main source data is a topographic map (Figure 3.1) and aerial photographs. Topographic map provide geographic reference which store the positional information of terrain features such as buildings, roads, rivers, contours, land use, administrative boundaries etc. Aerial photographs provide information about the city model textures. Several different types of software are employed to create the 3D models. ArcGIS was used to develop the foundation of their 3D city models while Canoma was used to construct a photorealistic building model. 3D Studio Max was used to authoring and integrating the foundation, building and other urban features (e.g. tree, lamp post and signboard). Cosmo player was used to visualise the 3D city models in X3D format. The workflow of model creation in Eran et al. (2005) is shown in Figure 3.2. and is described as follows. 1. DEM was generated from topographic map and aerial photograph using

ArcGIS software. 2. Superimposing or dragging a 2D representation of topographic features, e.g.

roads, rivers & land use information from orthophoto onto a DEM. 3. Create the initial 3D buildings based on the footprints and the heights from

aerial photograph. Then merge these models with the generated DEM. 4. Reconstruct photo-textured building models with multiple view point of

terrestrial images. 5. Edit the texture of building model is to eliminate the shadow effect and

obstructing objects like tree and vehicle.

27

6. Integrate both foundation of the initial city model with the photorealistic building models and adding-on extra accessories (landscape objects) such as trees, lamp post, sign board, bus stop etc.

Figure 3. 1: A topographic map (source: Eran et al., 2005)

Figure 3. 2: Workflow scheme of X3D model generation

(source: Eran et al., 2005)

28

The created models are saved in both X3D and VRML format and visualised with X3D viewers such as Cosmo player or Flux player. Since there are many viewers developed based on these standards. These standards can be used to present 3D city models but not represent them. The semantic information is missing, and it is difficult to integrate data from different standard format and make spatial analysis because of missing of semantic information.

3.1.3 Visualisation of CityGML CityGML as an OGC standard to represent 3D city models can be used to integrate both geometric and semantic information of the city models. Reitz et al. (2009) talked about the integration of CityGML and X3D for spatial data visualisation of Web 3D service. Figure 3.3 outlines architecture their Web 3D service. The geodata in CityGML format is first processed and stored in the Data Description, a structure containing meta-information about the object in the dataset (e.g. an object’s spatial location or its level of detail). Then the data is filtered according to certain rules and becomes the Filtered Data. Finally, the Portrayal Service creates the view, which consists of the filtered objects in their representation defined by Styled Layer Descriptors. Finally the Web 3D service renders this view into the exchange format requested by the client (for example VRML or X3D).

Figure 3. 3: The architecture of a web 3D service (source: Reitz et al., 2009)

In Figure 3.3, geospatial data in CityGML is filtered in two steps. The view will then be created according to Styled Layer Descriptors and encoded into a scene graph format like X3D before it is rendered to the client. (Reitz et al., 2009) Döllner et al. (2006) introduced the concepts, implementation, and experience of the Berlin virtual 3D city model. Figure 3.4 gives the overall architecture of Berlin 3D City Model System which is constituted by the following principal subsystems:

29

3D Authoring System: It is responsible for creating, editing, and versioning of the virtual 3D city models and its components, e.g., importing, exporting, grouping, and annotating buildings, vegetation plans, landscape plans, etc. Technically, it provides an interactive access to the 3D geo-database. 3D Geo-Database System: The database for storing and managing virtual 3D city models is based on the logical structure of CityGML. From Figure 3.4, it is clear that the CityGML based 3D Geo-Database System is the core of the whole system. Geo Data from different resources such as Cadastral Data, Digital Terrain Model, Aerial Photography, Building Models etc. is all converted into CityGML data and is integrated in a uniform database. With the CityGML format, exchange of 3D city models becomes much easier for different applications. And for visualisation, it is not necessary to consider variety of data sources but only focus on present the CityGML. 3D Presentation System: The presentation system provides real-time visualisation of and interaction with the virtual 3D city model. The 3D city models are converted into both KML and X3D format for different viewers. It is possible to access the Virtual Berlin through KML viewer like Google Earth and X3D viewer like LandXplore.

Figure 3. 4: The system architecture of the virtual 3D city model of Berlin

(source: Döllner et al., 2006)

30

Compared with other visualisation frameworks, it is reasonable to represent 3D city data in CityGML and visualise it with X3D or other 3D visualisation standards. Considering the connection between CityGML and X3D (Park, 2010), X3D is selected to visualise the 3D city models in this thesis.

31

3.2 Dynamic generalisation and visualisation City itself is so complex that it is always required to generalise the city data for effective visualisation. It is not only unnecessary to show every detail of the 3D city models to the user but also impossible for the user client to handle such huge amount of data. Visualising mass data actually is a disturbing factor preventing users retrieve their interested information from the 3D city models. In this thesis, two generalisation operators, aggregation and typification, are studied and a tree structure, CityTree, based on proposed visualisation framework is created for dynamic generalisation.

3.2.1 Aggregation A number of aggregation methods have been developed for aggregation of 2D cartographic objects and these methods can be also used in 3D. Bundy et al. (1995) introduced two types of aggregation operators Direct-merge operator and Snap-merge operator. The direct-merge operator maintains the alignment of the objects by moving the objects together directly. The relationship between the facing edges is represented by the triangles that have an edge in one object and a point in the other. Figure 3.5 gives an example of direct-merge. Meanwhile, the snap-merge tries to align the objects’ nearest parallel edges. It can be achieved by aligning the merge vector with the shortest outer connecting edge. Figure 3.6 shows the process of snap merge.

Figure 3. 5: Direct merge (Source: Bundy et al., 1995)

32

Figure 3. 6: Snap merge (source: Bundy et al., 1995)

Anders introduces an approach for 3D building aggregation using the following steps (Anders, 2005). 1) Compute the minimal bounding box of the building group using all points of the building geometries. 2) The length, width, and height of the bounding box define the three projection-directions L, W and H. 3) Create the orthogonal projection along height (Figure 3.7a), width (Figure 3.7b), and length (Figure 3.7c). 4) Generalise the three projections (Figure 3.8b). 5) Extrude the 2D geometries of orthogonal projection along height, width, and length respectively (Figure 3.8c). 6) The generalised 3D building group is computed by the intersection of the extruded geometries (Figure 3.8d). From the experiments, Andres’ 3D building aggregation method is quite suitable for buildings in a straight line with same orientation. Also their projections along height, width and length should also be in a straight line or superposition. These constrains may restrict the application of this method.

33

(a) Ground projection (b) Side projection

(c) Front projection

Figure 3. 7: 2D building projections (source: Anders, 2005)

(a) 3D Building group (b) Generalised projections

(c) 3D extrusion of the generalised

projections (d) Intersection of the extruded

projections Figure 3. 8: Aggregation of 3D buildings (source: Anders, 2005)

In Paper II, a novel aggregation method is proposed. Figure 3.9 gives a demo of the aggregation method. The aggregated ground plans will be also simplified to reduce the data volume. Compared with the above-mentioned aggregation algorithms, proposed method preserves the location of original ground plans, and creates the new edges by prolong the existing ones.

34

Figure 3. 9: Aggregation of ground plans

3.2.2 Dynamic typification Besides aggregation, typification is also an important generalisation operation for many city areas with regular building distribution. Typification is a generalisation operation that replaces a large number of objects by a smaller number of objects while trying to preserve the typical spatial structure of the objects. As discussed in Section 2.3.1, two types of typification methods, typification with and without structural knowledge are introduced in this section. Typification without structural knowledge Müller and Wang (1992) made use of mathematical morphology and proposed a solution to the automated generalise one particular class of geographical objects. Those objects are area patches (or ground plan polygons) distributed over a two-dimensional space, and are assumed to be no semantically difference among them. Certain rules are applied to typify the objects instead of detection the

35

structures. (1) emphasis on larger patches at the expense of smaller ones, (2) preservation of the overall figure/group relationship, (3) Partial topological integrity, and (4) differential displacement according to patch area. Even if they try to preserver the group relationship, it is quite hard task without the knowledge of the overall distributions. Sester and Brenner (2000) give another typification algorithm without structure knowledge. Their approach is based on Kohonen Feature nets (Kohonen, 1982), a neural network learning technique. The neurons of the nets have a property of keeping their spatial ordering when they are adapted to a new situation, which could be used in typification. The process is described as follows (Figure 3.10). First, determine the reduction rate using Töpfer’s law (Töpfer and Pillewizer, 1966). Then, random select a set of object as remain set, and create the Delaunay Triangulation to represent the topology of the objects in remain set. Next, introduce remain objects as the output objects of Kohonen net and go to learning phase. In the learning phase, the original objects act as attractors that drag the objects of the output map into their direction. In the learning iterations, the output objects are iteratively adjusted according to the underlying attractor structure. Finally, output objects are distributed across the space according to the original spatial distribution.

(a) (b) (c)

(d) (e) (f)

Figure 3. 10: Typification based on Kohonen Net (source: Sester, 2005)

36

In certain cases such as objects can be assumed as equal points, this approach produces reasonable results. However, in the 3D city models, objects like buildings are different from each other. Also the typification algorithms without structure knowledge can not preserve the dominant structures, like linear or grid, which are quite important in visualisation according to the Gestalt psychology. Therefore, the structure based typification algorithms are proposed. Typification with structural knowledge To generalise the city models with structure, first we have to detect the structure of the models. In the real city, there are several “regular” areas. By regular we mean that the buildings in these areas have some common property or structures e.g. linear structure in which buildings are evenly distributed along a road or grid structure lattice-like layout of buildings. Regnauld used minimum spanning tree (MST) to detect the linear structure and recognize the building clusters for generalisation (Regnauld, 1996). Each building is viewed as a forced cluster. The MST is composed with buildings as nodes and segments that represent the distance between two buildings as edges. The minimal distance between contours instead of centroids is taken. The MST is built by linking each building with its closest neighbor. Then the MST is segmented according two criteria: delete the long edges and make the buildings of a well-known structure to be part of a same linear cluster. Christophe and Ruas (2002) proposed a linear structure detection method based on straight line. This method can be processed by following two steps. First, identify the aligned buildings within a city block as the candidates for linear structure by scanning the whole space with straight lines. Second, qualify the generated candidates according to their regularities and eliminate the ones not regular. The drawback of this method is the computation load. Because it wants to scan every possible straight line in the space, if the interval is too small, it takes too long to calculate and if the interval is too big, some of the linear structures may be missed. Anders (2005) introduced an algorithm to detect the grid structure of 3D building groups. In a grid structure every building belongs to two linear structures which have to be preserved if possible. He uses relative neighbourhood graph (Toussaint, 1980) to detect the grid structure. The detected irregular grids then are abstracted to perfect regular grid with m x n members. The orientation of the grid is determined by the minimal enclosing rectangle (Freeman and Shapira, 1975). Then the abstracted regular grid structure is typified by reduce the number of rows and columns. For other non-structured part, the typification methods without structure knowledge can be employed.

37

Considering the linear structures are more common in cities and the grid structure can also be converted into linear structures. Therefore, the linear structure is selected for typification in this thesis. The methodology proposed in this thesis is given in Figure 3.11. First, the 3D city models are converted into CityGML LOD1 with only ground plan and height. Then, a minimum spanning tree (MST) is generated based on the distance between ground plan contours. Next, divide the created MST by the roads into sub-MSTs representing the groups of city models. Then lists of buildings suitable for typification are separated from the sub-MSTs. For each selected building list, typification is performed and saved for later dynamic visualisation. In the visualisation step, the invisible nodes in each building list are removed based on the coordinate of current view point or typification requirement. The remaining buildings in the list are adjusted in location and size. Finally, the city model is visualised using X3D. More information about the dynamic typification can be found in Paper IV. Compared with existing algorithms,

Figure 3.11: Work flow chat of dynamic typification

3.2.3 Multiple representation data structure With the help of above generalised methods, 3D city models in different LODs will be created. It is essential to integrate these data for effectively visualisation. Therefore, multiple representation data structure is required. The problem of handling multiple levels of details (LODs) in city models has been acknowledged for a long time (e.g. Köninger and Bartel, 1998) and the CityGML standard also specifies five LODs (OGC, 2008). In Paper II, some

38

related works of multiple representation data structure of 3D city models are introduced. Several existing structures are discussed in this section. For 2D objects, different tree structures are employed to represent the different map data in multi-LODs or multi-scales (Oosterom and Schenkelaars, 1995). The Binary Line Generalisation tree (BLG-tree) is used to represent the polyline. BLG-tree is created based on Douglas-Peucker algorithm (Douglas and Peucker, 1973). Reactive-tree based on R-tree is used to represent the any geometric type, e.g. point, polyline or polygon in the node form of (MBR, value, id), where MBR is the minimal axes-parallel bounding rectangle, imp-value is a positive integer indicating the importance level decided by type, attributes, size etc. of the object, and object-id contains a reference to the object. Every node can have children which represent the sub objects in its MBR. Generalised Area Partitioning (GAP)-tree is used to represent the hierarchy structure of area partitioning, which more accurately reflects the structure of the map than MBR used in reactive-tree. For 3D city models, based on R-tree, Coors (2003) proposes Progressive Tree or P-Tree to represent the hierarchical LOD of 3D city models. Compared with R-tree, P-tree has additional elements to store graphical information in this entry. This graphical information can be used to represent all child objects. Minimum bounding box is used to present the aggregation of all children. Figure 3.12 gives a P-tree structure of 3D city models.

Figure 3. 12: Example of P-tree (source: Coors, 2003)

3.3 Generalisation evaluation

39

As discussed in Section 2.3.2, evaluation of generalisation results is important. This section describes about some existing methods for generalisation evaluation and makes a comparison with proposed evaluation method using ARG and NEMD. Filippovska et al. (2009) proposed a framework to evaluate the quality of ground plan generalisation. They first classify the quality of generalised data on four different geometric aspects: trueness of ground plan, trueness of location, trueness of extension and trueness of shape. Trueness of ground plan can be measured by the largest deviation between the contour of the original and generalised ground plan. Trueness of location can be measured by the translation of the centroid before and after generalisation. Trueness of extension, the area distribution of an object in space can be measured by area difference, oriented bounding rectangles etc. Trueness of shape is a combined property which considers both the outline of a polygon and its area. The geometric properties such as compactness, roundness and convexity are suggested to measure the trueness of shape. In implementation, the trueness of ground plan is evaluated by boundary similarity Rboundary, the percentage of the remaining part of the polygon. Rboundary can be calculated with following function. Buffer(O) is a buffer around the original ground plan polygon which makes the matching of the two contours will not be based on strict geometric conditions, but rather tolerates small inaccuracies. Perimeter(G) and Perimeter(O) are the length of the contours of generalised and original ground plans.

( ) ( )

( )boundary

Buffer O Perimeter GR

Perimeter O

∩= (3.1 Source: Filippovska et al., 2009)

The trueness of extension and shape are measured by the ratio of intrusions (Rintrusion=Area(O∩G)/Area(O)) , extrusions (Rextrusion=Area(O∩G)/Area(G)) and area (Rarea=Area(G)/Area(O)). The total quality characteristic TQ is the Euclidean distance which aggregated all quality parameter like the rate of the boundary similarity, the rate of intrusions and extrusions, and the area ratio. Area and perimeter are also used by other studies (Podolskaya et al., 2007) to evaluate the quality of generalisation. However, these features can not reflect the similarity of city objects at certain generalisation operations like exaggeration or zooming. Bard (2004) proposed a more general quality assessment model based on: (1) characterisation of the data in their initial and final states at different levels of analysis; (2) a data quality assessment by comparison of the two characterisations; and (3) aggregation of the various assessment results to

40

summarize data quality. He divides the assessment functions into three parts characterisation functions, evaluation functions and aggregation functions. Characterisation functions try to describe the feature of object under assessment by geographic properties (geometric, semantic and topologic), such as area, or perimeter used in above ground plan evaluation method. These geographic properties can be at several levels of analysis micro, meso and macro (Ruas, 2000). The micro level represents individual features independent with other features such as size, height of a building. The meso level is made of groups of micro or meso features, such as a set of buildings (a micro feature) or a group of urban blocks (meso features). And the macro level is the overall information such as the mean height of all buildings. The evaluation functions are used to measure the similarity of certain feature between two objects. Three kinds of functions are presented: (1) threshold respect functions which check if the assessed property is over or under a threshold of legibility; (2) property evolution analysis functions which assess if a property has evolved as expected; and (3) qualification functions which arm to make evaluation results comparable. The aggregation functions are used to get overall quality of a generalisation result. It is composed of intra-class and inter-class aggregation. Intra-class aggregation is the assessment values for features of the same class such as building, while the inter-class aggregation is the aggregation of assessment results between different classes such as road and building. Classical operators of aggregation are the mean, the median, the minimum and the maximum. Also the weight set is quit important especially in inter-class aggregation.

41

4 Results and Discussions

4.1 Introduction In this chapter, the main results of the listed four papers are summarized and the contribution of each study to the knowledge in the field is also discussed. The papers are connected as follows. Paper I describes the basic framework of 3D city model visualisation that is used in Paper II and Paper IV. Paper II introduces a multiple 3D city data representation structure for more efficient visualisation based on the aggregation method. The quality evaluation is discussed in Paper III which can be used to assess the results of generalisation algorithms. In Paper IV, linear distributed objects of 3D city models are typified using Minimum Spanning Tree structure. 4.2 Visualisation framework The object of the study in Paper I is to develop a framework for 3D city model visualisation. This framework should be able to integrate data from different sources, to support urban 3D analysis such as model generalisation, user interaction etc. and to illustrate 3D scenes via Internet. Meanwhile, as a research prototype, this framework should also be easily extended and implemented. Therefore, CityGML is chosen as the data source because it is the 3D city model standard of OGC; Java is used for 3D analysis because it is widely supported by open source projects and is easily extended and implemented; X3D is selected for 3D visualisation because it is the 3D scene standard of the Internet and supports the real time user interaction. The proposed framework for 3DCM visualisation in Paper I is shown in Figure 4.1. Data in CityGML format is parsed and converted into Java classes to represent city objects such as buildings, roads etc. These classes may contain both geometry and semantic information (CityGML4j, 2010). Then based on the geometry or semantic information, 3D scenes of the 3D city models are generated. The created 3D scenes are then viewed on a standard web browser that supports Java Applet (Xj3D, 2010).

42

Figure 4.1: Framework of the online 3D city system

The input data set is in CityGML format, and the output visualisation results are in X3D. It is quite easy to implement and very suitable for research applications such as study of 3D model generalisation, spatial data analysis and results demonstration. Since CityGML and X3D are both XML based, it is possible to convert between them using Extensible Stylesheet Language Transformations (XSLT, 2010). This framework is more suitable than XSLT based visualisation for 3D city model related applications, because of difference between the geometry representations in CityGML and X3D, and the visualisation of semantic information, complex calculates are required. It is very difficult to create a XSLT to support such complex conversion. Therefore, the standard java classes are chosen in the framework. The key part of the visualisation, transform from CityGML to X3D is also discussed in Paper I. All LODs in CityGML are represented by GML multi-surface set which can be directly converted into X3D surface set object. While some other X3D elements are more suitable for lower LODs (LOD 0,1). LOD0 is a two and a half dimensional Digital Terrain Model, which could be converted to some other specific objects in X3D. In Paper I, the visualisation of texture is also discussed and the visualisation of 3D city model in CityGML through Internet is implemented as well using Xj3D related Java Applet. Visualisation 3D city models with CityGML and X3D are not new. The contribution of this study is mainly on the implementation side. The proposed framework is proved efficient and can be easily implemented and extended. Based on this framework, any API in java can be integrated for city data analysis. For example, in Paper II and IV, JTS is integrated to the framework and used to handle the calculation related to polygons, lines etc.

43

4.3 Multiple representation data structure Since the data volume for 3D city models usually is huge, 3D model generalisation is necessary to reduce the data volume and highlight the interested parts for more efficient visualisation. Meanwhile, 3D city model visualisation in different scales of the models is also required by many applications such as navigation, planning etc. Therefore, multiple representation data structure of the city models is necessary for continuous scaling and real time visualisation. The research object of this study is to create a multiple representation data structure of the city models with generalisation method to support continuous scaling. Based on the visualisation framework introduced in Paper I, CityTree, a multiple representation data structure of the city objects, is created to effectively realize the continuous scaling and dramatically reduce the load time of 3D models. By utilizing CityTree, it is possible to have dynamic zoom functionality in real time. When the 3D city models are in different LODs generalised, it is required to first convert the model from higher LODs to lower LODs. Here, nearby buildings are aggregated together to simplify the 3D city models, therefore all buildings are converted to LOD1 with only ground plan and height information in our implementation. A binary tree based structure CityTree is proposed for representation of multiple LODs data. Figure 4.2 gives a demo example of the CityTree. Figure 4.2(a) shows the distribution of the original city building group which contains 5 buildings (1~5). The rectangle areas (A~D) are created by selecting nearby buildings. The CityTree is generated as shown in Figure 3.4(b). The leaf nodes (1~5) are original objects in city model. The other nodes (A~D) are new generated middle nodes with geometry feature of aggregation of children models.

(a) (b)

Figure 4. 2: CityTree model

44

In the visualisation step, the CityTree nodes are selected based on the user’s view point and the features of the node. If the father node is selected to visualise, all his children will not be visualised which can reduce the computation complexity dramatically. The details about how CityTree is generated and visualised are given in Paper II. Compared with existing Reactive-tree GAP-tree or P-tree that are generated by dividing the area top-down, CityTree is easily implemented because it created by merging the nearest two nodes iteratively bottom-up. Also the geometry model of new generated node can represent its children properly because it is aggregation model of its children. Figure 4.3 describes the work flow of the CityTree creation. The ground plans of the buildings are generated and simplified. Then, the buildings are divided into clusters by the road. For each building cluster, a CityTree is created for visualisation. These four steps are the pre-process. With the CityTree, X3D scene will be created to represent the selected buildings in certain scale.

Figure 4.3: Work flow chat of generalisation

The ground plan of 3D building is generated by projecting its every surface into horizontal plane and merging them together. If the generated ground plan is not a single polygon, then save the model as multi-buildings which each polygon represent one building. Then simplify each ground plan polygon with simplification algorithm proposed by Sester and Brenner (2004) and modified by Fang (2009). Next create the Delaunay triangulation with each centroid of ground plans, cut the triangulation by the road and cluster the still connected

45

buildings into building groups. For each building group, create a CityTree which will be used for dynamic visualisation. The main contribution of this paper is the proposed framework of visualisation based on multiple representation data structure (CityTree) and generalisation methods to generate the different LOD representations. Delaunay triangulation and road network are used to clustering the concerted city objects e.g. buildings in this paper. An improved ground plan aggregation algorithm is proposed using merging edge selection and increasing area modification by keeping the basic location of original ground plans and simplifying the aggregated ground plan to reduce the data volume. Based on the clustering and aggregation, CityTree, a multiple representation data structure of the city objects, is created and used to visualise the 3D city models of multi-scale dynamically.

4.4 Generalisation results evaluation Paper III is focused on typification of the building facade structure or windows. The contribution of the author of this thesis is mainly about the generalisation evaluation method based on Attributed Relationship Graph (ARG) and Nested Earth Mover’s Distance algorithm (NEMD) (Kim et al., 2004). In order to evaluate the results of generalisation, ARG and NEMD methods are employed in this paper. Most existing frameworks for results evaluation are based on the generalisation algorithm. In other word, the corresponding between original and generalised objects has to be known, which unfortunately in most cases isn’t. Therefore, a novel generalisation quality evaluation algorithm using ARG and NEMD is proposed in this thesis. The proposed method takes original and generalised models as input without knowing their correspondence. Then the ARG of inputted models are generated to represent the feature of the objects in the models (micro features) and the relationship between these objects (meso features). Next, NEMD algorithm is employed to calculate the correspondence of the original and generalised objects. Finally the overall generalisation quality is assessed based on the “distance” between the corresponding objects. The detail of the algorithm is introduced in Paper 4, in which the visual similarity is the main criteria for the quality assessment. An ARG G is defined as G = {V, R} in which V={v i|1<=i<=n} and R={r ij|1<=i<=n, 1<=j<=n}.V is the set of n nodes and each vi represents a building. R is n*n matrix and each r ij is the relationship between the building vi and vj. In this application the node will contains the features about the building such as area and height, and the relationship between nodes will represent the spatial and topological relations between the buildings. The ARG matching can

46

be implemented as a two-step procedure, first constructing a distance matrix and second establishing the correspondence based on the distance matrix. Kim et al. (2004) present the NEMD algorithm using the nested structure of earth mover’s distance (EMD). More specifically, the nested structure of the EMD consists of inner EMD and outer EMD: The inner EMD reflects the difference of both nodes and edges in two ARG’s, and the outer EMD establishes the correspondence between nodes in the two ARG’s. This paper focuses on ARG of building group generation and constructing the distance matrix of nodes and relations for similarity evaluation. Meanwhile, the evaluation method also certifies the effectiveness of the proposed typification algorithm. The contribution of this study is that the quality evaluation method of visual similarity calculation is proven to be efficient according the user survey results. This evaluation method makes use of ARG for feature representation of 3D city models and NEMD for difference calculation, which is novel in the generalisation quality assessment field.

4.5 Typification based on MST Paper IV is mainly about typification of 3D city models. In a real city, there are several “regular” areas which are suitable for typification. By regular we mean that the buildings in these areas have some common property e.g. evenly distributed along a road. The suitable generalisation method for these regular areas is typification that replaces a large number of objects by a smaller number of objects while trying to ensure that the typical spatial structure of the objects is preserved (Shea and McMaster, 1989). In this paper, the linear structure of city object is adopted for typification. The typification methodology proposed in this thesis is given in Figure 4.4. The chains of buildings suitable for typification are selected using minimum spanning tree (MST). For each chain, the typification method is removing the building with smallest ground plan area. Then adjust the positions of remain buildings along the building chain to keep their initial distance ratio and extend the model of remaining building to the same sum area of the initial buildings.

47

Ground plan generation & simplification

Preprocess in

Java

environment

Run in X3D

MST creation

Model visualization

Building chain creation

Typification of building groups

Figure 4.4. Work flow chat of dynamic typification.

For each selected building list, typification is performed and the results are saved for later dynamic visualisation. In the visualisation step, the number of removed nodes in each building list is decided based on the coordinate of current view point or typification requirement. The remaining buildings in the list are adjusted in location and size. Finally, the city model is visualised using X3D. The contribution of this study is the typification algorithm. The proposed algorithm for typification is original and it has the advantage to be very simple and efficient. It often gives acceptable results within a specific scale range. Meanwhile, the liner structure widely exists in the urban area, so the proposed method can be applied for 3D city model generalisation extensively.

48

49

5 Conclusions and future research

5.1 Conclusions This thesis focuses on the visualisation and generalisation of 3D city models in multi scales. A visualisation framework is proposed based on CityGML and X3D to represent and present the 3D city models respectively. Considering the importance of generalisation to the effective visualisation, two common used generalisation operators, aggregation and typification are studied. A multiple representation data structure, CityTree, is proposed to support dynamic aggregation and visualisation. Minimum Spanning Tree is used to identify the city areas with linear structure buildings and these areas will be typified with different strategies. The results of typification are evaluated by first generating the ARGs to represent their features and then matching these ARGs with the ARG of the original models with NEMD algorithm. The experiments results certify the following contributions of this thesis which answer to the scientific questions proposed in section 1.2. 1) How to represent and visualise 3D city model efficiently? The proposed framework can implement the visualisation of 3D city models both offline and online. The framework takes CityGML files as input, analyses the 3D city models in Java platform and presents 3D scene using X3D related technologies. It is possible to represent and integrate data from different sources in CityGML, to make use of existing Java analysis source codes and to visualise the 3D scene by software or browsers that support X3D. The proposed framework can represent and visualise 3D city model efficiently. (2) How to automatically generate 3D city models in different LODs and implement dynamic visualisation of 3D city models in multi-scale? Aggregation and typification methods are studied in this thesis to generalise the 3D city models and to create the models in different LODs automatically. A multiple representation data structure, CityTree is proposed to dynamically visualise the aggregated models in different LODs and reduce the load time dramatically. Minimum spanning tree (MST) based typification algorithm is selected for generalisation of city models with linear building structures along the roads. (3) How to assess the visual quality of generalised 3D city models?

50

The proposed evaluation method based on ARG and NEMD can correctly reflect the visual percept similarity between generalised models and the original ones according to the user survey. Through the study of this thesis, the specific objectives proposed in section 1.2 are achieved. First, make use of the 3D city model standard CityGML and create online and offline 3D scenes with X3D. Second, study current cartographic generalisation methods and propose new algorithms (Dynamic aggregation and MST based typification) to support 3D city model generalisation. Third, find out the suitable structure (ARG) to represent the visual feature of 3D city models, and evaluate the similarity between generalised models and original ones using NEMD algorithm.

5.2 Future research Future research is planned to tackle a number of potential research issues related to this study. First is generalisation of textured 3D city models. Texture is extremely important to create virtual reality environment and is often required in different 3D city related applications for better visualisation effects. Current studies are mainly focus on the geometry generalisation, but few of them are on texture which actually is a quit important part and takes high percentages of the total data volume of 3D city models. Second, not only linear structure but also other structures like grid should be considered for typification. It is required to identify more “regular” areas in the city, and find suitable methods to typify them. Meanwhile, it is also important to integrate different methods such as typification and aggregation for an entire city model generalisation and to generate the unified models in different LODs. Third, the evaluation algorithm should become more conform to the human visual perceptions. User survey should be carried out and their similarity evaluation should be used to improve the accuracy of the proposed evaluation method. Besides, new technologies such as HTML5 should be studied to create more efficient 3D online visualisation. Even though the online 3D visualisation of X3D can be implemented using Java Applet, the load time is quite long and it is hard to support huge 3D models. The HTML5 related techniques such as X3DOM (Behr et al., 2009) maybe a better choice for online 3D presentation. We should continually improve the framework by making use of the new coming techniques.

51

References 3DCityDBv2, 2009. Available online at: http://opportunity.bv.tu-

berlin.de/software/projects/show/3dcitydb, (accessed 22 Feb 2010) 3DMLW, 2010. http://www.3dmlw.com/, (accessed 22 Feb 2010) Alessandro, M., Nadalutti D., and Chittaro L., 2007. Interactive Walkthrough of

Large 3D Models of Buildings on Mobile Devices, in Web3D 2007: Perugia, Italy.

Anders, K., 2005. Level of detail generation of 3D building groups by aggregation and typification, in Proceedings of XXII International Cartographic Conference 2005, La Coruna, Spain

Anders, K. and Sester M., 2000. Parameter-Free Cluster Detection in Spatial Databases and its Application to Typification. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 33: pp. 75-82

Apache Pivot, 2010. http://pivot.apache.org/, (accessed 22 Feb 2010) Araujo, B., Iwasaki M., Pizzolato B., and Boukerche A., 2008. Framework for

3D Web-Based Visualization of HLA-Compliant Simulations, Proc. 13th Int. Symp. on 3D Web Technology (Web3D 08), ACM, Los Angeles, California, 2008

Aristoteles, 2009. Available online at: http://www.ikg.uni-bonn.de/aristoteles/ index.php/Aristoteles, (accessed 22 Feb 2010)

Ban, Y. F. et al., 2008. ViSuCity, http://www.infra.kth.se/GIS/research/projects/ ViSuCity/, (accessed 22 Feb 2010)

Bard, S., 2004. Quality Assessment of Cartographic Generalisation. Transactions in GIS, 8(1): pp. 63–81.

Beck, M., 2003. Real-Time Visualisation of big 3D City Models, International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol.34-5/W10, pp. 1-6

Bederson, B.B. and Hollan, J.D., 1994. Pad++: A Zoomable Graphical Interface. Computer-human interaction conference (CHI'94), short paper.

Behr, J., Eschler P., Jung Y. and Zöllner M., 2009. X3DOM: a DOM-based HTML5/X3D integration model, Web3D '09, Darmstadt, Germany

Bell, G., Parisi A. and Pesce M., 1995. The Virtual Reality Modeling Language-Version 1.0 Specification, Available online at: http://www.web3d.org/ technicalinfo/specifications /VRML1.0/index.html /, (accessed 22 Feb 2010)

Bilasco, I. M., Marlène V., Gensel J. and Hervé M., 2007. Semantic-based Rules for 3D Scene Adaptation, in Web3D’07: Perugia, Italy.

BIMServer, 2010. Available online at: http://www.bimserver.org/, (accessed 22 Feb 2010)

Bit Management Contact viewer, 2010. http://www.bitmanagement.com/, (accessed 22 Feb 2010)

52

BS Contact Geo 7.2, 2009. Available online at: http://www.bitmanagement. com/developer/contact/relnotes72Geo.html, (accessed 22 Feb 2010)

BS Editor, 2010. http://www.bitmanagement.de/en/products/authoring-tools/ bs-editor, (accessed 22 Feb 2010)

Bundy, L., Jones B., and Furse E., 1995. Holistic Generalisation of Large-Scale Cartographic Data. In: Müller, J. C., Lagrange, J. P., and Weibel, R. (Eds.), GIS and Generalisation, Gisdata 1, Taylor & Francis, pp. 106-119.

Burghard, D. and Cecconi A., 2007. Mesh simplification for building typification. International Journal of Geographical Information Science, 21(3): pp. 283-298

Buttussi, F., Chittaro L. and Coppo M., 2007. Using Web3D Technologies for Visualization and Search of Signs in an International Sign Language Dictionary, in Web3D 2007: Perugia, Italy.

Cabral, M. et al., 2007. An experience using X3D for Virtual Cultural Heritage, in Proc. of 12th International Conference on 3D Web Technology (Web3D 07), Perugia, Italy, pp161-164

Calado, A., Abreu P., Chumbinho, R., Silva, A., Sousa, L. and EMEPC Group, 2008. Integration of Hydrographic Data Products in a Global Web Based 2D and 3D GIS, Proceedings of the Canadian Hydrographic Conference and National Surveyors Conference 2008

Campbell, A., 2007. Building Information Modeling: The Web3D Application for AEC, in Proc. of 12th International Conference on 3D Web Technology (Web3D 07), Perugia, Italy, pp173-176

Carlson, W., 2003. A Critical History of Computer Graphics and Animation, Available online at: http://design.osu.edu/carlson/history/lessons.html, (accessed 22 Feb 2010)

ChefX3D, 2007. http://www.chefx3d.org/index.html, (accessed 22 Feb 2010) Chen, A., Leptoukh G., Kempler S., Nadeau D. and Zhang Z., 2008.

Augmenting the research value of geospatial data using Google Earth. In: De Paor, D. (Ed.), Google Earth Science Journal of the Virtual Explorer, electronic edition, vol. 29, Paper100

Christophe, S. and Ruas A., 2002. Detecting Building Alignments for Generalisation Purposes, in International Symposium on Spatial Data Handling, Ottawa, Canada. pp. 419-432

CityGML4j, 2010. Available online at: http://opportunity.bv.tu-berlin.de/software/projects/show/citygml4j, (accessed 22 Feb 2010)

CityVu, 2009. Available online at: http://cityvu.3dgis.it/lang/en/, (accessed 22 Feb 2010)

COLLADA, 2010. https://collada.org/mediawiki/index.php/COLLADA_-_Digital_Asset_and_FX_Exchange_Schema

Coors, V., 2003. 3D-GIS in networking environments, Computers, Environment and Urban Systems Volume 27, Issue 4, July 2003, Pages 345-357

53

Courtin, O. and Jonglez D., 2007. A full Web 3D GIS, using PostGIS and X3D output, FOSS4G2007

Czerwinski, A., Sandmann S., Stöcker-Meier E. and Plümer L., 2007. Sustainable SDI for EU noise mapping in NRW – best practice for INSPIRE, International Journal of Spatial Dana Infrastructures Research, No. 2, 90-111, 2007.

Czerwinski, A., Kolbe T., Plümer L. and Stöcker-Meier E., 2006a. Interoperability and accuracy requirements for EU environmental noise mapping, in International Symposium InterCarto-InterGIS. 2006: Berlin.

Czerwinski, A., Kolbe T., Plümer L. and Stöcker-Meier E., 2006b. Spatial data infrastructure techniques for flexible noise mapping strategies, in 20th International Conference on Informatics for Environmental Protection EnviroInfo. 2006: Graz.

Dehghan, S. and Steel R., 1997. Small cell city, IEEE Commun. Mag., vol. 35, pp. 52-59, Aug. 1997.

Deng, F., Zhang Z. and Zhang J., 1995. Construct 3D City Model by Multi-sensor Data, ISPRS Workshop on Service and Application of Spatial Data Infrastructure, XXXVI (4/W6), pp 187-190, Oct.14-16, Hangzhou, China

Digital Nautical Chart, 2010. http://www.nga.mil/portal/site/dnc/, (accessed 22 Feb 2010)

DirectX, 2010. http://www.microsoft.com/games/en-US/aboutGFW/pages/ directx.aspx, (accessed 22 Feb 2010)

Döllner, J., Kolbe T., Liecke F., Sgouros T. and Teichmann K., 2006. The Virtual 3D City Model of Berlin - Managing, Integrating and Communicating Complex Urban Information, in 25th International Symposium on Urban Data Management UDMS. 2006: Aal.

Döllner, J. and Hagedorn B., 2007. Integrating Urban GIS, CAD, and BIM Data by Service-Based Virtual 3D City Models. 26th Urban Data Management Symposium (UDMS 2007),Stuttgart, Germany, October, 2007

Douglads, H., and Peucketr K., 1973. Algorithms for the reduction of points required to represent a digitized line or its caricature, Canadian Cartographer, Vol. 10, pp. 112- 122.

Eastman, C., 2009. What is BIM? Available online at: http://bim.arch.gatech. edu/?id=402, (accessed 22 Feb 2010)

Eliens, A., Wang Y., Riel C. and Scholte T., 2007. 3D digital dossiers: a new way of presenting cultural heritage on the web, in Proc. of 12th International Conference on 3D Web Technology (Web3D 07), Perugia, Italy, pp157-160

Eran, S. et al., 2005. The Design and Development of a Virtual 3D City Model. Available online at: http://www.hitl.washington.edu/people/bdc/ virtualcities.pdf, (accessed 22 Feb 2010)

ETU VDI 3805 Navigator, 2008. Available online at: http://www.ifcwiki.org/ index.php/ETU_VDI_3805_Navigator_IFC_Export, (accessed 22 Feb 2010)

54

Fan, H. and Meng L., 2010. An automatic approach for the typification of facade structures, in Submitted to 13th AGILE International Conference on Geographic Information Science, Portugal

Fan, H., Meng L. and Jahnke M., 2009. Generalization of 3D buildings modelled by CityGML. In: Lecture Notes in Geoinformation and Cartography, Advances in GIScience. Springer Heidelberg Berlin, ISSN 1863-2246, pp 387-405, 12th AGILE International Conference on Geographic Information Science, Hannover, Germany

Filippovska, Y., Kada M., Fritsch D., 2009. Quality Evaluation of Ground Plan Generalization, 12th AGILE International Conference on Geographic Information Science 2009, Hannover, Germany

Flux, 2007. http://mediamachines.wordpress.com/, (accessed 22 Feb 2010) Flux Studio, 2007. http://mediamachines.wordpress.com/flux-player-and-flux-

studio/, (accessed 22 Feb 2010) FME 2010 Data Inspector, 2009. Available online at:

http://evangelism.safe.com/fmeevangelist29/, (accessed 22 Feb 2010) Forberg, A., 2007. Generalization of 3D building data based on a scale-space

approach, ISPRS Journal of Photogrammetry and Remote Sensing, 62(2): pp. 104-111

Freeman, H. and Shapira R., 1975. Determining the minimum-area encasing rectangle for an arbitrary closed curve, Communications of the ACM. Vol. 18, 409-413.

Frueh, C., 2002. Automated 3D Model Generation for Urban Environments, Ph.D. Thesis, University of Karlsruhe.

Frueh, C. and Zakhor, A., 2003. Constructing 3D city models by merging ground-based and airborne views, IEEE Computer Graphics and Applications, Special Issue Nov/Dlec. pp. 52-41.

FZKViewer 0.9, 2009. Available online at: http://www.iai.fzk.de/www-extern/index.php?id=1975&L=1, (accessed 22 Feb 2010)

Glander, T. and Döllner J., 2007. Cell-Based Generalization of 3D Building Groups with Outlier Management, Proceedings of the 15th International Symposium on Advances in Geographic Information Systems, ACMGIS 2007

Glide, 2010. http://glide.sourceforge.net/, (accessed 22 Feb 2010) Google, 2008. We know the web is big…, http://googleblog.blogspot.com

/2008/07/we-knew-web-was-big.html (accessed 22 Feb 2010) Götzelmann, T., Guercke R., Brenner C. and Sester M., 2009. Terrain-

dependent Aggregation of 3D City Models, ISPRS workshop in Lund, December 2009

Hamza-Lup, G., Davis L., and Zeidan O., 2006. Web-Based 3D Planning Tool for Radiation Therapy Treatment, 11th International Symposium on 3D Web Technology (Web3D 06), 18–21 April, Columbia, Maryland

55

Harrie, L., 2001. An Optimisation Approach to Cartographic Generalization, Doctor Thesis, Lund University

Havele, A., 2010. Web3D and OGC to build cohesive standards for location-based 3D Web visualization services, http://www.web3d.org/press, (accessed 22 Feb 2010)

Hermsmeyer, D. and Guretzki M., 2005. A 3d City Model of Kuwait: Data Processing and Possible Applications, Map Middle East 2005, http://www.gisdevelopment.net/technology/gis/me05_137pf.htm, (accessed 22 Feb 2010)

HTML5, 2010. http://dev.w3.org/html5/spec/Overview.html, (accessed 22 Feb 2010)

IFC, 2009. Available online at: http://www.ifcwiki.org/index.php/Main_Page, (accessed 22 Feb 2010)

IfcExplorer, 2007. Available online at: http://www.ifcwiki.org/index.php/ IfcExplorer_CityGML_Export, (accessed 22 Feb 2010)

InstantReality, 2010. http://www.instantreality.org/home/, (accessed 22 Feb 2010)

Internet World Stats, 2009. The Internet Big Picture, Available online at: http://www.internetworldstats.com/stats.htm (accessed 22 Feb 2010)

JavaFX, 2010.: http://javafx.com/, (accessed 22 Feb 2010) Jeospace, 2010. http://www.jeospace.org/, (accessed 22 Feb 2010) Jung, Y. and Behr J., 2008. Extending H-Anim and X3D for Advanced

Animation Control, in Web3D 2008: L.A. USA. Jung, Y., Recker R., Olbrich M., and Bockholt U., 2008. Using X3D for medical

training simulations, Proc. 13th Int. Symp. in 3D Web Technology (Web3D 08), Los Angeles, California, 2008, pp. 43-51

Jung, Y., 2007. Enhancing X3D for advanced MR appliances, in Web3D 2007: Perugia, Italy

Kada, M., 2002. Automatic generalization of 3D building models. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 34(4): pp. 243-248

Kada, M., 2006. 3D Building Generalization based on Half-Space Modeling, in Proceedings of the ISPRS Workshop on Multiple Representation and Interoperability of Spatial Data, Hannover.

Kada, M., 2007. Generalisation of 3D Building Models by Cell Decomposition and Primitive Instancing. In: Proceedings of the Joint ISPRS Workshop on“Visualization and Exploration of Geospatial Data” , Stuttgart, Germany.

Kada, M., 2009. The 3D Berlin Project, Photogrammetric Week 2009, Stuttgart, Germany from 7-11 September 2009. 331-340

Kanellos, M., 2005. New Life for Moore's Law. Available online at: http://news. cnet.com/New-life-for-Moores-Law/2009-1006_3-5672485.html (accessed 22 Feb 2010)

56

Kim, H., Yun D. and Lee S., 2004. A New Attributed Relational Graph Matching Algorithm using the Nested Structure of Earth Mover's Distance. In Proceedings of 17th International Conference on Pattern Recognition (ICPR'04).

KML, 2010. http://www.opengeospatial.org/standards/kml/, (accessed 22 Feb 2010)

Kohonen, T., 1982. Self-organized formation of topologically correct feature maps. Biological Cybernetics, 43:59-69.

Kolbe, T. H. and Gröger G., 2003. Towards unified 3D city models, Proceedings of the ISPRS Commission IV Joint Workshop on Challenges in Geospatial Analysis, Integration and Visualization II , Stuttgart.

Kolbe, T. H., Gröger, G., and Plümer, L., 2005. CityGML -- Interoperable Access to 3D City Models. In Proceedings of the First International Symposium on Geo-Information for Disaster Management (Delft, Netherlands, March 21--23, 2005)

Kolbe, T. H., 2007. CityGML–3D Geospatial and Semantic Modelling of Urban Structures, Presentation on the GITA / OGC Emerging Technologies Summit in Washington, online available at http://www.citygml.org/ fileadmin/citygml/docs/CityGML_ETS4_2007-03-21.pdf, (accessed 22 Feb 2010)

Kolbe, T. H., 2008. Representing and exchanging 3D city models with CityGML. In: Zlatanova, S., and Lee, J. (Eds.), 3D Geo-Information Sciences, Springer, pp. 15-31.

Köninger, A., and Bartel S., 1998. 3D-GIS for Urban Purposes. GeoInformatica, Vol. 2, No. 1, pp. 79-103.

LandXPlorer CityGML Viewer 2009a, 2009. Available online at: http://usa.autodesk.com/adsk/servlet/pc/index?id=12561246&siteID=12312, (accessed 22 Feb 2010)

Lerma, J. L. and Garcia A., 2004. 3D City Modelling and Visualization of Historyical Centers, International Workshop on Vision Techniques applied to the Rehabilitation of City Centers. 25-27 October, Lisbon, Portugal

Mackaness, W., Ruas A., and Sarjakoski L. T., 2007. Generalisation of Geographic Information: Cartographic Modelling and Applications. Elsevier

Mayer, H., 2005. Scale-spaces for generalization of 3D buildings, In: International Journal of Geographical Information Science. Vol. 19, No. 8-9, September-October 2005, pp. 975-997

MBARI, (2010): http://www.mbari.org/about/, (accessed 22 Feb 2010) Meng, L. and Forberg A., 2007. 3D Building generalisation, In: W. Mackaness,

Ruas A., and Sarjakoski L. T. (eds.), Generalisation of Geographic Information: Cartographic Modelling and Applications, Elsevier, pp. 211-232.

Mesa 3D, 2010. http://www.mesa3d.org/, (accessed 22 Feb 2010)

57

Microsoft Silverlight, 2010. http://www.microsoft.com/SILVERLIGHT/, (accessed 22 Feb 2010)

Ming, Y., J. Jiang and F. Bian, 2002. 3D-City Model Supporting for CCTV Monitoring System, Symposium on Geospatial Theory Processing and Application, Ottawa 2002

Moore, G., 1965. Cramming more components onto integrated circuits, Electronics, Vol.39, pp. 1–4.

Moritz, N., Dirk B., and Robert W., 2009. Automated processing for map generalization using web services. Geoinformatica (Dordrecht), Vol. 13(4): pp. 425-452

Müller, J. and Wang Z., 1992. Area-patch generalization: A competitive approach. The Cartographic Journal, 29(2):137–144

Murata, M., 2004. 3D-GIS Application for Urban Planning based on 3D City Model, ESRI User Conference Proceeding, August 9-13, 2004

Nielsen, J., 1998. Nielsen's Law of Internet Bandwidth, Available online at: http://www.useit.com/alertbox/980405.html (accessed 22 Feb 2010)

Nurminen, A., 2006. m-LOMA - a Mobile 3D City Map, in Web3D 2006: Columbia, Maryland.

Nurminen, A., 2007. Mobile, hardware-accelerated urban 3D maps in 3G networks, in Web3D 2007: Perugia, Italy.

O3D, 2010. http://code.google.com/apis/o3d/, (accessed 22 Feb 2010) Octaga, 2010. http://www.octaga.com/, (accessed 22 Feb 2010) OGC, 2008. City Geography Markup Language (CityGML) Encoding Standard,

Available online at: http://www.opengeospatial.org/standards/citygml, (accessed 22 Feb 2010)

Olsson, E., 2009. Exporting 3D Geoinformation from Baggis Database to CityGML, KTH master thesis, TRITA-GIT EX 09-05. Supervisor: Yifang Ban. April 2009.

Oosterom, P. V., and Schenkelaars V., 1995. The Development of an Interactive Multi-Scale GIS. International Journal of Geographical Information Systems, Vol. 9, No. 5, pp. 489-507.

OpenGL, 2010. http://www.opengl.org/, (accessed 22 Feb 2010) Pairs3D, 2009. Available online at: http://itunes.apple.com/WebObjects/

MZStore.woa/wa/ viewSoftware?id=323822225&mt=8 (accessed 22 Feb 2010)

Pantzer, J., 2008. Dusseldorf-traffic simulation with municipal CityGML geodata, 24th Plenary Meeting of the SIG 3D, Available online at: http://www.citygml.org/fileadmin/citygml/docs/2008-06-03_4_CityGML_D_sseldorf.pdf, (accessed 22 Feb 2010)

Park, M., 2010. Web3D and OGC to Build Cohesive Standards for Location-Based 3D Web Visualisation Services, available at http://www.prweb.com/ releases/2010/02/prweb3482634.htm. (accessed 25 March 2010).

58

Patterson, D., 2007. 3D SPACE: Using Depth and Movement for Selection Tasks, in Web3D 2007: Perugia, Italy.

Pittarello, F. and Faveri A., 2006. Semantic Description of 3D Environments: a Proposal Based on Web Standards, in Web3D 2006: Columbia, Maryland.

Planet 9 Virtual Cities, 2010. http://www.planet9.com/products_cities/ products_virtcity.html, (accessed 22 Feb 2010)

Podolskaya, E., Anders K., Haunert J. and Sester M., 2007. Quality Assessment for Polygon Generalization, 5th International Symposium on Spatial Data Quality (SDQ 2007), ITC, Enschede, The Netherlands.

Ptolemy3D, 2010. Available online at: http://www.ptolemy3d.org/, (accessed 22 Feb 2010)

Regnauld, N., 1996. Recognition of Building Clusters for Generalization, in Proceedings of the 7th Spatial Data Handling Symposium: Delft, the Netherlands. pp. 185-198

Reitz, T., Kramer M., Thum S., 2009. A processing pipeline for X3D earth-based spatial data view services. Proc. 14th International Conference on 3D Web Technology, Darmstadt, Germany, 16-17 June, pp. 137-145

Rez, 2007. http://www.rez3d.com/, (accessed 22 Feb 2010) Rich, S., 2009. Available online at: http://www.spatialexplorations.net/blog/

2009/8/22/the-need-for-comprehensive-3d-city-models-part-1.html, (accessed 22 Feb 2010)

RISpec, 2010. http://renderman.pixar.com/products/rispec/, (accessed 22 Feb 2010)

Ruas, A., 2000. The role of meso object for generalisation. In Proceedings of the Ninth International Symposium on Spatial Data Handling, Beijing, China: 3b.50–63

Sadek, E., Jamaludin A., Rosdi B. and Kadzim M., 2002. The Design and Development of a Virtual 3D City Model, Available online at: http://www.hitl.washington.edu/people/bdc/virtualcities.pdf, (accessed 22 Feb 2010)

Schulte, C. and Coors V., 2008. Development of a CityGML ADE for dynamic 3D flood information, In Proceedings Joint ISCRAM-CHINA and GI4DM Conference on Information Systems for Crisis Management, Harbin, China.

Schulze-Horsel, M., 2007. 3D City Models:Data Generation and Applications, Available online at: http://www.gisdevelopment.net/magazine/years/2007/ april/26_1.htm (accessed 22 Feb 2010)

Sester, M., 2000. Generalization based on least squares adjustment. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 33(4): pp. 931-938

Sester, M., 2005. Optimization approaches for generalization and data abstraction, International Journal of Geographical Information Science, Vol. 19, No. 8–9, pp. 871-897

59

Sester, M. and Brenner C., 2000. Typification based on Kohonen Feature Nets, in Presented in the 1st International Conference on Geographic Information Science (GIScience 2000) Savannah, Georgia, USA

Sester, M. and Brenner C., 2004. Continuous Generalization for fast and smooth Visualization on small Displays. International Archives of Photogrammetry and Remote Sensing, vol 34, part 4, pp. 1293-1298

Shea, S., and McMaster R., 1989. Cartographic Generalization in a Digital Environment: When and How to Generalization, Proceedings of 9th International Symposium on Computer-Assisted Cartography, Baltimore, pp. 56-67

Shinozaki, M., 2008, Virtual 3D Models in Urban Design, Virtual Geographic Environment 2008, Hong Kong

SwirlX3D Editor, 2010. www.pinecoast.com/swirl3d.htm, (accessed 22 Feb 2010)

Tang, L, Chen C., Huang H. and Lin K., 2006. Research on HLA-based Forest Fire Fighting Simulation, the 9th AGILE Conference on Geographic Information Science, Visegrád, Hungary, 2006 System

Thiemann, F., 2002. Generalization of 3D building data. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 34(4): pp. 286-290

Thomas, B. J., 2008. WebTOP: An X3D-Based, Web-Delivered, Interactive System for Optics Instruction, in Web3D 2008, L.A. USA.

Toussaint, T. 1980. The Relative Neighbourhood Graph of a Finite Planar Set, Pattern Recognition, Vol. 12, pp. 261-268

Töpfer, F. and Pillewizer W., 1966. The Principles of Selection, The Cartographic Journal, Vol. 3, No. 1, pp. 10-16

U3D, (2010): http://www.ecma-international.org/publications/standards/Ecma-363.htm, (accessed 22 Feb 2010)

United Nations, (2008): World urbanization prospects: the 2007 revision. Highlights, New York: Population Division, Department of Economic and Social Affairs, United Nations. Available at: http://www.un.org/esa/ population/publications/wup2007/2007WUP_Highlights_web.pdf. , (accessed 22 Feb 2010)

VDI 3805, 2010. Available online at: http://www.vdi3805.org/, (accessed 22 Feb 2010) , (accessed 22 Feb 2010)

Verma V., Rakesh H., 2006. 3D Building Detection and Modeling from Aerial LIDAR Data, Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’06)

VirturalGL, 2010. http://www.virtualgl.org/, (accessed 22 Feb 2010) Vivaty Studio, 2010. http://www.vivaty.com/downloads/studio/, (accessed 22

Feb 2010) Weber, C. and Parisi T., 2007. An Open Protocol for Wide-area Multi-user X3D,

in Web3D 2007: Perugia, Italy.

60

Weibel, R. and Dutton G., 1999. Generalizing Spatial Data and Dealing with Multiple Representations. In: Longley, P., Goodchild, M.F., Maguire, D.J. and Rhind, D.W. (eds.): Geographical Information Systems: Principles, Techniques, Management and Applications, Second Edition. publisher: John Wiley, 125-155.

Willis, S., 2007. Protein CorreLogo: an X3D representation of co-evolving pairs, tertiary structure, ligand binding pockets and protein-protein interactions in protein families, in Proc. of 12th International Conference on 3D Web Technology (Web3D 07), Perugia, Italy, pp. 71-80.

WorldWind, 2008. http://worldwind.arc.nasa.gov/java/, (accessed 22 Feb 2010) X3D, 2010. http://www.web3d.org/x3d/, (accessed 22 Feb 2010) X3D Earth, 2010. http://www.web3d.org/x3d-earth/, (accessed 22 Feb 2010) X3D-Editor, 2010. https://savage.nps.edu/X3D-Edit/, (accessed 22 Feb 2010) Xj3D, 2010. http://www.xj3d.org/, (accessed 22 Feb 2010) XSLT, 2010. http://www.w3.org/TR/xslt#section-Introduction, (accessed 22 Feb

2010) Yamagishi, Y., et al., 2010. Visualisation of geoscience data on Google Earth:

Development of a data converter system for seismic tomographic models. Computers & Geosciences 36 (3), 373-382.

Zhang, X. P., Yan T, Cui X. and Zhu D., 2009. Quantitatively Optimized Selection of Proper Scale in Land Use Cartographic Generalization, IFIP International Federation for Information Processing, Vol. 293/2009, Springer Boston, ISSN:1571-5736 (Print) 1861-2288 (Online)

Zhang, X., Ai T. and Stoter J., 2008. The Evaluation of Spatial Distribution Density in Map Generalization, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B2. Beijing

Zhu, A., Li F. and Zhang Y., 1995. Unified Representation of Three Dimensional City Models, ISPRS Workshop on Service and Application of Spatial Data Infrastructure, XXXVI(4/W6), Oct.14-16, Hangzhou, China