Researh On Traffic Light Recognition Method Based On A Car-like Robot

In recent years, with the highlight of traffic safety problems, the research on the smart car has become a hot topic. However, considering the current level of domestic research on intelligent vehicle, there are still some difficulties in verifying the cognitive of intelligent vehicle and realizing the multi-car interaction. While micro-smart car has a great advantage in low costs of research and production, short development cycle, small running space and more safety, which just solves problems of intelligent vehicle in a high proportion of development costs, the experimental space limitations, security and other issues.This paper focuses on the cognitive problem of microfilm smart car to traffic lights in the micro-traffic environment. For the traffic light detection, an effective recognition method is proposed. The proposed algorithm includes two parts: traffic-light color detection and shape recognition. In traffic light color detection, a ROI is predefined for eliminating background information as well as reducing time consuming. This is helpful for efficiency improvement.By the channel independence of HSV color space, traffic light candidate regions are extracted based on statistical principles. For traffic light shape recognition, by analyzing the attributes of the traffic lights in micro-traffic environment, three shape features are extracted from candidate regions of traffic light. The used extraction process is pretty simple, which can ensure a car-like robot recognize a traffic light in real-time. In order to achieve high recognition accuracy, SVM classifier is used based the extracted shape features.Finally, many off-line experiments and online tests on the car-like robot have been done.The experimental results prove that the proposed method can identify the traffic lights well and achieve a high accuracy, and is able to satisfy the requirements of real-time for the car-like robot.