DETECTING AND RECOGNIZING TRAFFIC LIGHTS BY GENETIC APPROXIMATE ELLIPSE DETECTION AND SPATIAL TEXTURE LAYOUTS

作者：Chiang Cheng Chin^*; Ho Ming Che; Liao Hong Sheng; Pratama Andi; Syu Wei Cheng

来源：International Journal of Innovative Computing Information and Control, 2011, 7(12): 6919-6934.

• Traffic light recognition
• Color segmentation
• Genetic algorithm
• Approximate ellipse detection
• Local binary pattern (LBP)
• Support vector machine

Traffic light detection is usually treated as a circle detection problem in computer vision. To handle more challenging cases in traffic light detection, this paper extends the circle detection problem to an approximate ellipse detection problem. For tackling the approximate ellipse detection problem, we propose a novel genetic approach which is more robust in handling appearance changes caused by perspective shape deformations and partial occlusions. To deal with the color recognition of each detected traffic light, we propose the design of a spatial texture layout feature which is more effective in handling illumination variations under different weather conditions and eliminating false alarms from irrelevant scene backgrounds. The experimental results show that the proposed method achieves an average recognition rate of 95.01%, with a false alarm rate below 2% based on 763 3-color traffic lights over 714 testing images, and demonstrates superior performance compared with an existing method.

• 出版日期2011-12
• 单位东华大学