Research On Vehicle Behavioral Trajectory Detection Technology Based On Video Image

With the improvement of traffic management and traffic monitoring intelligent level, video surveillance technologies based on video image processing, analysis and understanding have drawn increasing attention recently. Among them, the detection technology based on video image is a method to obtain the vehicle information effectively and quickly. It can monitor the movement of the vehicle, such as random stop, retrograde, or speeding. It not only can provide legal basis for traffic monitoring, but also can provide data support for the analysis of traffic conditions, which make the traffic more intelligent, more convenient and more efficient. This paper is based on the binocular vision technology to expand the study of the trajectory of object, including the moving target detection and tracking, the camera calibration, stereo matching, and moving object three-dimensional trajectory generation and so on. Among which it focuses on the detection and tracking moving object, and stereo matching algorithm.Firstly, in the detection and tracking of moving vehicles, the LK sparse optical flow algorithm is introduced in this paper. Aiming at the problem that the target detection and tracking is not accurate in the traditional LK optical flow algorithm, this paper proposes a new automatic detection and tracking algorithm for multiple moving targets with PRLK and GMM, and the template update mechanism is introduced to deal with the phenomenon of turning and deformation in the process of moving vehicles. The whole process can be carried out without supervision of automatic detection and tracking, and the algorithm is tested under three different traffic scenarios. Experiments show that the algorithm can track the moving vehicle for a long time, and also can adapt to the change of the target scale. At the same time, the processing time data of traffic video is collected to verify the real-time performance of the algorithm. The processing speed is 24.58fps,28.65fps and 27.36fps respectively, which can meet the needs of real-time traffic.Secondly, the Zhang Zhengyou board calibration method is introduced, which is based on the traditional calibration and self-calibration, and the camera calibration is completed by using the Matlab stereo calibration kit, and the parameters of the camera and the stereo vision system are obtained. Then the target object is tracked by two cameras and the 3D trajectory is reconstructed and restored by triangulation method. In this paper, we propose an improved stereo matching algorithm based on SURF feature, and propose a matching of feature points by using KNN-RANSAC algorithm. Experiments show that this method is effective and can eliminate more false matching points.