Real Time Vehicle Detection On Road

Autonomous Land Vehicle is an intelligent robot which can move on road or field automatically, which has great value on researches and applications. This paper focuses on one of the technologies of ALV, autonomous vehicle detection on free ways with one camera.Firstly, the paper studies on the method of extracting the suspected vehicle area according to the vehicle shadows. A simple method is presented to determine the threshold of the binary image by estimating the gray values of the road surface. Compared with the improved histogram method and edge segment method, this method has more adaptability and higher efficiency.Secondly,the merits of the symmetry measurement in the vehicle detection have been specifically analyzed and five constraint conditions of vehicle areas are presented. Using these conditions, a new method is presented to verify vehicle areas and locate the position of vehicles accurately at the same time. The experiments show that this method has low false alarm rate.Basing on the above detection method for single frames, a policy using information of serial images is presented, which decreases the detection time by 20 percents, furthermore, the false alarm rate and undetected error rate are also decreased effectively.The experiments on more than 1000 real images show that our vehicle detection system can detect and track multi vehicles in real time and the performance of detection system is robust in complicated environments.