Research And Implementation Of Rear Approaching Vehicle Detection Algorithm Based On Fish Eye Camera

With the accelerated urbanization process and the development of social economy, the pressure of transportation system is increasing. Therefore, the importance of intelligent transportation systems is growing, which is based on computer vision technology. The core technology of that is the vehicle detection and recognition, which are based on the analysis and processing of the camera’s image sequence. The thesis presents a rear approaching vehicle detection algorithm, which is based on fish eye camera. Experimental results show that the method has a high detection performance. The main content are summarized below.First, the algorithm makes distortion correction of the images obtained by fish eye camera to eliminate the impact of that to the algorithm. Then the algorithm uses frames subtraction method to select vehicle candidate area to determine the region where vehicles may exist. Next, the thesis propose a method of regional average feature detection to concentrate the features on the vehicles. Then by judging the characteristics of the optical flow vector, the areas of vehicles are identified initially. At the same time, the algorithm uses the method of reversing tracking feature points and the method of optical flow cluster to reduce the errors in tracking feature points. Finally, for the judgment conditions of vehicle optical are too simple, in order to determine the final vehicle area, the thesis uses the training classification algorithm of Adaboost to identify the vehicle region again. At the stage, in order to achieve the best training effect, the thesis makes the study of the choice of samples and the appropriate size of them. And propose a new training termination condition, which can effectively prevent training time from being too long and avoids an uncompleted training.Firstly, this paper introduces the vehicle detection applications and the development status of that. Secondly a new algorithm which combines the method of frame subtraction and optical flow is proposed. Thirdly, we use the algorithm of Adaboost to obtain the finally vehicle areas. At last, the performance of the algorithm is analyzed. It’s proved experimentally that the algorithm is robust.