Research On Moving Object Tracking Algrithm In Vedio

Moving object tracking in video assembles advanced technologies in many fields such as image processing, pattern recognition, artificial intelligence, automatic control computer application, physiology, physics and mathematics, et al. It has many excellent advantages of visual scene, anti-interference and cost-effective, and has been widely applied to military surveillance, traffic control, machine intelligence, medical diagnosis, and so on. With the continuous demand of complex environment applications, besides introducing new technology, how to improve the accuracy and robust of the existing tracking algorithm is the focus of the current work.In this thesis, we describe the method that how the MeanShift algorithm applied in object tracking field, and weight the location information on the color histogram. By analyzing results of MeanShift algorithm tracking experiment, we get the inherent defects of MeanShift Algorithm. The main work of the thesis is present a new method to improve two defects of MeanShift Algorithm.First, for the problem that the MeanShift algorithm cannot track object that is occlusion completely, we propose a solution that combined MeanShift algorithm and Kalman filter, which use Kalman filter to predict the object position in current frame, then search the target in the neighborhood using MeanShift algorithm. In the thesis, we divide the occlusion status into three statements: no occlusion, partly occlusion, completely occlusion. In addition, the residual of Kalman is the criteria of three statuses. When the object is occlusive completely, we predict its location by its linear motion features. For dynamic template, this paper employed Case-based Reason method. It show good performance which the fixed template cannot handle.Then, we employed Continuously Adaptive MeanShift (CamShift) algorithm, which can resize kernel window adaptively. Using HSV color model improves the algorithm performance on optical-sensitive. Using ellipse searching window makes the search result of transforming object more accurate. At last, this thesis introduces an automatic object tracking strategy, which effectively reduces the errors that caused by initial the target manually.