Reseach On Video Tracjing Of Pedesteains In Natural Scene

Motion detection and tracking of the human body is an important research field in computer vision, which also is the basic in the analysis and understanding of human behavior. Detection and tracking in pedestrian have two kinds of video sequence which are static background and moving background, this study is based on static single camera motion tracking. Before carrying out tracking, it needs target detection and division for video images of moving objects we are interested in, and then using the image processing and feature recognition. This paper mainly be summarized as follows:Firstly, we introduce the development of target tracking technology and analyze several mature motion target method, then we compare different methods of application of occasions. Based on the background subtraction method, the problem of various scenarios for the natural complexity, this paper has a comprehensive analysis of advantages and disadvantages in the different Gaussian modeling of object detection algorithms in the pedestrian. Because of the vulnerability which is updated slow and poor convergence, we have researched the algorithm for Gaussian mixture background model that based on adaptive learning rate, and do some experiments for target detection with this algorithm.When all the moving target are detected out from the videos, then we use a variety of image processing techniques to extract complete targets that this paper are interested in, moreover, using the characteristics which are different from other objects to identify pedestrian target. Firstly, using mathematical morphology analysis to eliminate the specific noise; then using regional connectivity technology to fill the hole of the binary image. What’s more, with the characteristics that human eye’s ability to distinguish details of the color resolving power is lower than the details of intensity resolving power, we transformat the RGB color space to the HSV color space, we have experiments to suppress and eliminate the shadow of the target partition, in the end using the pedestrian’s own characteristics to identify the desired target we need.In pedestrian tracking, we firstly analysis the status of the traditional Kalman filtering in tracking field, in connection with the application of Kalman filtering and then we study the feasibility of the extended Kalman filter algorithm which is in the fast tracking. And then we focus on the particle filter in tracking the theory and application, based on the existing problem of particle degradation,we combine the iteration unscrented Kalman filter and particle filter tracking algorithm to track the pedestrian target, which has enriched the use of video tracking Knowledge, also enhanced the rapid and robust of the pedestrian tracking in different scenarios.