Algorithm Researches On Single Object Tracking For Video Surveillance System And Its Implementation On DSP System

With the rapid development of electronic information, as one of the crucial issues of computer vision, visual object tracking is widely used in many applications, such as visual surveillance, visual navigation of robots, human-computer interaction, medical diagnose and military guidance. After widely analysis of the typical algorithm of object tracking, Mean Shift algorithm was deeply studied based on hardware in the implementation of video surveillance. Mean Shift algorithm is a kind of fast and effective method, through the probability density gradient, with multiple iterations, to look for the object location. Mean Shift meets the real-time requirements, and has become an important content of video object tracking method. It is important that object tracking under complex conditions using the dedicated hardware.Based on the area of video surveillance applications using a single fixed camera, video object tracking is realized on hardware. The specific work is as follows:The thesis has introduced the basic principle of classical Mean Shift algorithm and its application in video object tracking, sumarized the design flow charts and the implementation steps, and used the Microsoft Visual C++ 6.0 to do the simulation of Mean Shift algorithm. Based on the simulation results, the advantages and disadvantages of Mean Shift algorithm are analysed to improve this algorithm.Part occlusion is studied, edge weighted Bhattacharyya coefficient is used to determine the time of occlusion and the template updating strategy.This thesis has adopted an improved algorithm based on object feature gray screening, in view of the traditional Mean Shift algorithm because of the background pixels in the deviation of target problem, the improved algorithm on the background of the object model and object model surrounding area modeling, respectively, to obtain the eigenvalues of the normalized gray-level histogram. If a feature is greater than the specified threshold value in the object model and less than a specified threshold, then such characteristics are considered as the pixel. The improved algorithm can effectively mitigate the effects of background pixels on the object positioning, improve the object tracking accuracy.In this thesis, the improved algorithm is implementaed on the hardware of ICETEK-DM6437-B evaluation board, the experimental results show that that the improved Mean Shift algorithm achieves good performance and meets the requirements of real-time and high effiecncy. It could be easily realized to get accurate tracking position for low-speed targets.