Design And Implementation On Moving Object Detection And Tracking Algorithm In Real-time Monitoring

With the rapid improvement of science and technology, especially the development of sensor and its fusion, camera can be seen everywhere in people’ s daily life. Target detection and tracking in the video is to get the various mobile information at any time for people’s interest, which is the basis for many of the related fields of study.Moving target detection is mainly used to extract the moving ta rget in video through the corresponding algorithm, and it is the foundation of the subsequent target tracking. With tracking the moving targets, all kinds of information of targets can be obtained, including the direction of movement, the speed, the accele ration, the color and the center of the target. Tracking the target accurately is the key to many computer vision research, target tracking can provide a variety of motion data for the behavior analysis, target recognition, human-computer interaction.This paper introduces some commonly used methods, for each algorithm, the writer pointed out its advantages and disadvantages through experiments.This paper has mainly studied the background difference method and frame difference method. In this paper, by combining the two algorithms, using the detected edge produced by adjacent frame difference method to connect the broken areas produced by background difference method, we can get the complete target. Due to the existence of shadows, the region with shadows could mistake the results of target detection, the paper has also used the corresponding algorithm to remove shadows.With the complete detected targets, this paper has designed a kind of detection and tracking algorithm based on the features of targets. Through comparing the features of targets got from the adjacent frames, we have finished target tracking, and the method is applied to traffic flow statistics, statistical accuracy rate reached 86%. We focus on the Mean Shift tracking algorithm which is based on the kernel function color histogram, with the advantages of small calculation quantity, not sensitive to target rotation, deformation of nonparametric density estimation, but the algorithm also has some disadvantages, such as similar color interference, fixed tracking window bandwidth, sensitive to Occlusion. In this paper, by using the block color histogram to solve the problem that Mean Shift is easily affected by the background color, combined with the Kalman filter to solve the target occlusion problem, using the tracking window width and length adaptive adjustment to solve the problem of the fixed bandwidth of Mean Shift tracking window.