California Car License Plate Recognition System ZhengHui Hu Advisor: Dr. Kang.
-
date post
21-Dec-2015 -
Category
Documents
-
view
213 -
download
0
Transcript of California Car License Plate Recognition System ZhengHui Hu Advisor: Dr. Kang.
California Car License Plate California Car License Plate Recognition SystemRecognition System
ZhengHui Hu
Advisor: Dr. Kang
12/11/2006 License Plate Recognition System 2
IntroductionIntroduction
A License Plate Recognition System (LPRS) is a system to automatically detect, recognize and identify a vehicle plate.
It involves low-level image processing techniques with higher level artificial intelligence techniques.
12/11/2006 License Plate Recognition System 3
ApplicationsApplications
Mainly for monitoring, surveillance and security. For example,– Entrance/Exit monitoring for parking lot
structures– Part of surveillance system for gated
communities– Control gateways for vehicle passage– Security Systems for high traffic
Law Enforcement
12/11/2006 License Plate Recognition System 4
Technical IssuesTechnical Issues
Image Capturing– Vehicle speed– Lighting condition– Occlusion
Processing speed– Heavy traffic
Recognition accuracy– High correctness
12/11/2006 License Plate Recognition System 5
Current StateCurrent State
There are many companies, especially in Europe, that developed this type of system commercially
There are many research trying to improve accuracy and speed performance
12/11/2006 License Plate Recognition System 6
System ArchitectureSystem Architecture
Plate Region Segmentation– Locate plate region out of car
and/or background Character Segmentation
– Segment each character/number out of plate
Character Recognition– Recognize each character on
the plate– Similar to OCR process
Plate Segmentation Module
Character Segmentation Module
Character Recognition Module
12/11/2006 License Plate Recognition System 7
Previous WorkPrevious Work
Most previous work are focused on the character segmentation and recognition process based on– Fuzzy algorithms– Template matching– Neural network
12/11/2006 License Plate Recognition System 9
Step1 - Plate Region SegmentationStep1 - Plate Region Segmentation
Goal– Locate license plate in an image
Target image group– California Car License Plates (regular ones)
Challenges– Location: plate regions at random place – Size: vehicle distance from the camera affect plate size– Color: affected by lighting conditions (day/night/shadow)– Skew/distortion: images can be taken from different angles
12/11/2006 License Plate Recognition System 10
Step1 - Plate Region SegmentationStep1 - Plate Region Segmentation
Helpful information– All License Plate have same shape– Known background/foreground colors
Light background color Bluish foreground color
– numbers and characters
– Color distribution in a rectangular plate region
12/11/2006 License Plate Recognition System 11
Step1 - Plate Region SegmentationStep1 - Plate Region Segmentation
Image Preprocessing
Filter using Color and Edge Information
Connected Components Analysis
Edge Information
Find Candidate Regions
Input Image
Feedback for more filtering
12/11/2006 License Plate Recognition System 12
Input ImagesInput Images
Captured using a digital camera – Different distance– Different lighting
conditions– Different angles
Original size 2048X1536
Resized to 800X600 for faster process
12/11/2006 License Plate Recognition System 14
Edge InformationEdge Information
Apply morphological operator to detect region of high change.
Plate character/numbers are among these
12/11/2006 License Plate Recognition System 15
FilterFilter
Filter using Color and Edge Information– Use edge information
to find plate background color
– Filter image using plate background color
12/11/2006 License Plate Recognition System 16
Connected Component AnalysisConnected Component Analysis
Find connected component and values– Width/Height ratio– Amount of edge pixels
12/11/2006 License Plate Recognition System 17
Find CandidateFind Candidate
Plate has ratio between 1 and 3
Plate has highest or 2nd highest pixel density from edge image
12/11/2006 License Plate Recognition System 18
Experiment Results Experiment Results
Total Pictures Tested: 43 – Region found: 38– Region not found: 5 – Success rate: 88%
Error classification– Filtering process chopped out part of plate– Fail to identify correct candidate region
12/11/2006 License Plate Recognition System 19
Experiment Results (Experiment Results (SpeedSpeed))
Machine Used for Testing: Pentium 4-M 1.70Ghz, 256 MB RAM– For images 800X600, the processing time is
150 ~ 190 ms– For original size image 2048X1536,
processing time is around 1 sec
12/11/2006 License Plate Recognition System 20
Step2 - Step2 - Character SegmentationCharacter Segmentation
Segment each character/number out of the plate detected by previous module
Challenges– Rectangle segmented might contain more
than just the plate– Plate might contain some things other than
number/characters
Still under development
12/11/2006 License Plate Recognition System 21
Step3 - Step3 - Character RecognitionCharacter Recognition
A process to recognize each character/number segmented
Challenges– Noise– Image scaling and distortion – Image corruption
12/11/2006 License Plate Recognition System 22
Step3 - Step3 - Character RecognitionCharacter Recognition
Our approach– Artificial Neural networks are used to
recognize characters and digits– During training process, simulated annealing
process was added to the back propagation training to avoid the problem of local minimal
Still under development
12/11/2006 License Plate Recognition System 23
ConclusionConclusion
Contributions – Algorithm for plate detection– Combination of back-propagation/simulated
annealing process in neural network trainingFuture Work
– Improve recognition ratio in step1 via feedback for filtering and better connected component analysis
– Finish Step3
12/11/2006 License Plate Recognition System 24
ReferencesReferences
See Resources page in Website for full list of references