Video-based Traffic Light Recognition Algorithm

Video-based Traffic light recognition algorithm is widely used in the smart car and auxiliary driving system. Combined with the current research, this paper presents real-time, effective and accurate recognition method to accommodate various shapes of traffic light. The improved algorithm adapts to varying illumination conditions and complex background environment, and experiments are carried out on the front camera of the electronic police system. The main work is as follows:1. Using the closed loop of image control system and image enhancement to ensure the quality of the input video sequences, then, down sampling and initial positioning in order to compress algorithm’s time and space expends.2.An improved HSV color segmentation method is proposed to adapt to the color error and overexposed. Use the improvement of connected region analysis contribute to enhance the speed and object detection, the geometric constraint model to filter to get the target region. Then,combined with relatively independent segmentation results of color and shape information,utilize color principal component judgment and multi-frame detection statistic to obtain the preliminary traffic light recognition result. Experimental data show that these improvements above have good effects.3.Proposes a PCA-HOG based SVM classify algorithm to recognize traffic lights, for the HOG descriptor’s excellent geometric rotation and light changing invariant.4.Completes and tests the algorithm on PC and camera of the electronic police system. The experiment results show this method accommodate multiple types of traffic lights, at the same time, it is real-time, accuracy and robust in complex environments and varying illumination conditions.