Introduction to OpenCV Dr. Chung-Hao Chen Haole Guo Sep 2011.
Chen, Hao Department of Transportation Engineering,
description
Transcript of Chen, Hao Department of Transportation Engineering,
© 2008 Chen Hao, Beijing Institute of Technology1
Intelligent Parking System: Parking Guide Application in Beijing and Method for
License Plate Localization
Chen, HaoDepartment of Transportation Engineering,
Beijing Institute of Technology, Beijing, PR China. [email protected]
http://chenhaoits.cn
2© 2008 Chen Hao, Beijing Institute of Technology
Basic Information Our lab
Department of Transportation Engineering School of Mechanical & Vehicular Engineering
• Four research teamsTraffic Plan and DesignTraffic Information and ManagementLogistic OperationAutomobile Application
Me• B.S., Transportation Engineering, Beijing Institute of Technology, China. 2005
Paper: Public Transportation Query System of Beijing • M.S., Transportation Engineering, Beijing Institute of Technology, China, 2007
Thesis: Research and Application of License Plate System
• Research Assistant in the Intelligent Transportation Lab• Research Interests:
Intelligent Transportation Systems, Image Processing and Pattern Recognition, GIS, Dynamic Traffic Assignment, Traffic Modeling and Simulation, Pavement Design.
3© 2008 Chen Hao, Beijing Institute of Technology
Outline: Parking Guide System A project of our lab (demonstrated by the government for 2008
Olympic Games in Beijing )
License Plate Recognition System Mainly introduce the plate localization process: correlation based method
• Brief Introduction• Candidate Area Extraction • Candidate Verification • Experimental results
4© 2008 Chen Hao, Beijing Institute of Technology
1. Parking Guide System• Circumstance
The parking lots of Cui Wei shopping mall in BeijingTwo floors, 267 parking spaces
• Functions:Automatic parking guideEasily Management for employees Alleviate traffic pressure on the road
• Three Key TechnologiesUltrasonic DetectorCAN Bus Communication LED Billboard
5© 2008 Chen Hao, Beijing Institute of Technology
Layer Components• Data Collection Layer
the basic layercollect the data from all parking spaces
• District Layergather the data from Parking Space Detectore.g. one District Collector connect with 60 detectors
• Floor Layerfloor layer controls all data from the corresponding district collectors
• Central Layerall data input to the central collectorinformation broadcasting
6© 2008 Chen Hao, Beijing Institute of Technology
Framework of Our Parking Guide System
Floor Controller
7© 2008 Chen Hao, Beijing Institute of Technology
8© 2008 Chen Hao, Beijing Institute of Technology
Parking Space Detector
Central Controller
District Controller
9© 2008 Chen Hao, Beijing Institute of Technology
2. License Plate Recognition System 2.1 Brief Introduction• Three parts
license plate localization candidate extraction candidate verification
character segmentationcharacter recognition
• Difficultiesweather, illuminationlicense plate: size, colorshoot anglepollution and abrasionshelter problem High quality camera and image processing board
(arithmetic embedded in a digital processing chip)
10© 2008 Chen Hao, Beijing Institute of Technology
plate localization
character normalization and recognition
character segmentation
An Example:
11© 2008 Chen Hao, Beijing Institute of Technology
2.2 Candidate Area Extraction
• Preprocessing and Rank Filter• Searching Reference Lines • Get candidate areas
(a)
(d) (e)
(b) (c)
12© 2008 Chen Hao, Beijing Institute of Technology
• Auto-Correlation Based Algorithm:
Auto-correlation algorithm; (a) calculate the auto-correlation
property; (b) auto-correlation curve;
Base on the characteristic that plate area has seven block areas.
The auto-correlation curve has about thirteen peaks for the car plate.
2.3 Candidate Verification
(a)(b)
13© 2008 Chen Hao, Beijing Institute of Technology
• Projection Based Algorithm:
Verify the car plate using projection
algorithmVerify the headlight
area using projection algorithm
14© 2008 Chen Hao, Beijing Institute of Technology
Framework of proposed candidate verification method. In our experiment, th1=7, th2=20, th3=5, th4=20.
15© 2008 Chen Hao, Beijing Institute of Technology
• Step 1: Extract plate which has light characters in dark background.
• Step 2: Extract plate which has dark characters in light background.
16© 2008 Chen Hao, Beijing Institute of Technology
• Step 3: Extract blue-white plate which has polluted by the light at night.
17© 2008 Chen Hao, Beijing Institute of Technology
2.4 Experimental results• Database
– 720*280 JPEG color images from a park entrance
– 1704 images from three days surveillance are tested
• Results– [1] : 88.1%
– Proposed: 97.5%
Algorithms Images Correct No Plate
[1] 1704 1501 188
Proposed 1704 1661 21
[1] V. Shapiro, G. Gluhchev, D. Dimov, Towards a multinational car license plate recognition system, Machine Vision Application, Volume 17, Issue 3, July 2006. pp. 173 – 183
18© 2008 Chen Hao, Beijing Institute of Technology
Thanks for your attention!
Q&A