Research On Detection And Tracking Algorithm Of Human Motion For Intelligent Video Surveillance

With the rapid growth of economic level and informatization of the modern society, the applications of computer vision analytical understanding become increasing in people’s daily life and work. Research direction covers the following categories:target detection and classification, target tracking and target recognition. The objects for researching fall into two categories:rigid object and non-rigid object. The detection and tracking algorithm in this paper is for human body, the algorithm is proposed improvements on the basis of existing algorithm. The main work in this paper includes two things:(1) This paper presents an improved three differential real-time detection algorithm for motion target. Current common detection algorithms have background subtraction, three differential method. We can know that the results of background subtraction have obvious "empty" and a part of noise from the experimental results in this paper. We also know that the target through the traditional three differential detection method obtained contains o lot of "empty" and "double shadow" phenomenon, so the experiment results are poor. Therefore, this paper proposed an improved three differential algorithm based on the traditional three differential method. The basic idea of the algorithm is to first select a effective background through Surendra algorithm and update the obtained background through selectively background updating method. Secondly we use the adjacent three images to make the difference with the updated background image, and the shadow which difference object contains is eliminated in HSV color space. Thirdly we make three differential method and the result make "and" operation. Finally we use the result after "and" operation and the intermediate frame image which obtain after background subtraction to make "or" operation, so we can get complete moving target.(2) This paper presents an improved algorithm for human target tracking based on Meanshift algorithm. When human body make obviously changes, the tracking box of traditional Meanshift method can not adaptive change and when the body is partially obscured, traditional Meanshift method is difficult to accurately track the human body and track occurs significant deviation. Therefore, the idea of the proposed algorithm in this paper is that taking the improved three difference method to extract moving target area and obtain the contour of target. At the same time, Meanshift method is to track the moving targets, get maximum probability area and use the center coordinate of the region as the center of current frame tracking box. We judge motion target is whether to be blocked through the value of Pap coefficient. If the target is blocked, we should use kalman filter to predict the path of moving target and the size of tracking frame is changed according to the size of moving target..In this paper, experiments of detection and tracking algorithm is achieved by programming on Matlab software platform and analysis the experimental results in detail. It verifies the effectiveness and correctness of the improved algorithm.