Traffic Sign Detection
Project: Autonomous Driving System
(Module: Traffic Sign Detection)
Supervisors : Professor Seiichi Mita, Dr. Son Tran
(Toyota Technological Institute of Nagoya, Japan)
Abstract. This paper presents a new approach for color detection and segmentation based on Support Vector Machine (SVM) to retrieve candidate regions of traffic signs in real-time video processing. Instead of processing on each pixel, this approach utilizes a block of pixels as an input vector of SVM for color classification, where the dimension of each vector can be extended by a group of neighboring pixels. This helps to handle the diversification of data on both training and testing samples. After that, Hough transform and contour detection are applied to verify the candidate regions by detecting shape of circle and triangle. The experimental results are highly accurate and robust on the testing database, where samples are recorded on various states of environment.
The color training dataset and testing dataset are published for free, more details are presented in the paper.
- Red color training dataset ~ 5MB
- Blue color training dataset ~ 12MB
- Testing dataset ~ 110MB (Thanks to Siman Ying, Beijing University of Technology, China, for keeping its backup)
* Download code: (backup link)
(Just for references, no warranties for any errors, please take care when using it)
(All of the code is C++, it requires OpenCV, SVM-Light)
Please using my original link and do not mirror anywhere. If you use my datasets or code, please cite the following paper:
Tam Le, Son Tran, Seichii Mita, Thuc Nguyen, Realtime Traffic Sign Detection Using Color and Shape-Based Features, The 2nd Asian Conference on Intelligent Information and Database Systems (ACIIDS), LNAI5991, Vietnam, 2010 .
Related publication: