Real-time traffic light detection on resource-limited mobile platform

Given the rapid expansion of car ownership worldwide, vehicle safety is an increasingly critical issue in the automobile industry. The reduced cost of intelligent mobile phones has made it economically feasible to develop intelligent systems for visual-based event detection for forward collision avoidance and mitigation. In this work, a real-time traffic red light recognition is proposed under mobile platforms. The proposed method consists of real-time traffic lights localization via image down-sampling, circular regions detection and further traffic lights recognition. Hough Transform is modified to fast localize the traffic light candidates. Finally, a strong classifier is made from multiple weak features is employed for further verifications. In the experiment, the detection rate can achieve above 70%. This shows that our proposed traffic light recognition can be applied in real world environments.