Research On Algorithm For Traffic Light Detection And Recognition

As traffic condition tends to be complicated, more and more Traffic lights are needed to provide navigation information at the corner for vehicles. The running intelligent vehicle must obtain traffic light information to make its behavior decision. This dissertation presents a real-time algorithm of detection and recognition for circle-shaped and arrow-shaped Traffic lights. To increase the detection speed, this algorithm firstly reduced the quality of the original image, and secondly used color and basic geometric characteristics to locate the Traffic light. Modeling the arrow-shaped Traffic light, use it and the color and basic geometric characteristics of traffic light, circular degree to detect Traffic light, thus derive the candidates. Fourthly, Gabor+2DPCA feature extraction algorithm was used to extract feature of candidates, which was recognized by template matching.The main work of this dissertation is as follows:1. As there is still no geometric model to describe the arrow-shaped Traffic light accurately, this dissertation models the arrow-shaped Traffic light by its marginal feature and region saturation.2. For the first time, Gabor+2DPCA feature extraction algorithm was used to extract the texture feature of the candidate in Traffic light recognition. This algorithm was compared with HU invariant moments feature extraction algorithms and Fourier descriptor feature extraction algorithms.3. Aiming at the problem of sample library redundancy, this dissertation designed an algorithm of autonomous library-building. It rejected the samples that are of little difference and derived the target sample library finally.The proposed algorithm was coded, and a system for Traffic light detection and recognition was designed. Experiments were carried out to analyze the performance of this system. The analysis shows that the proposed algorithm has higher recognition rate, meeting the real-time demand. For per-frame, the average recognition rate of circle-shaped Traffic light and arrow-shaped Traffic light is96.4%and95.1%, and the process time is48.4ms and98.2ms respectively.