Research Of Target Tracking Algorithm Based On Improved Mean Shift

It is an information world now. The rapid development of computer technology has brought great convenience to our work and life. Video target trackong is an interdisciplinary field based on computer and information technology, which is to find the most similar one with the specified target part in a video sequence.It has already been applied in video surveillance, human-machine interface, traffic surveillance and military,etc.In the subject research,by means of learning and exploring the issue of target tracking in sequence of video images,on the basis of summarizing the extensive research results of predecessors, the subject research has made some progress.Due to the complexity of the background and the environment of target tracking, it is difficult to find an algorithm that can fit all situations, so variety of algorithm have been proposed. This thesis mainly studies Mean Shift alrorithm,and simply analyzes Kalman filter and particle filter algorithm, including the basic idea of the two filtering algorithms and their advantages and disadvantages. Finally, describe and validate a moving target tracking algorithm proposed by predecessors, that is Mean Shift tracking algorithm based on comer points, since Harris corner detection is adopted in this algorithm. We call this algorithm Harris-Mean Shift algorithm. Against the insufficience of certain aspects of Harris-Mean Shift algorithm, an improved algorithm is proposed.Mean Shift algorithm is a modeling method based on the characteristics and statistics of probability density. In the tracking process, the target area is usually selected by the user in the first frame of the video sequence, and the appropriate target color histogram is established. According to Bhattacharyya similarity. Mean Shift algorithm iteratively search for the best candidate regional for the target model in subsequent frames. This method enables Mean Shift algorithm to have good performance in tracking, for example:good real-time, robustness on the occlusion and deformation. But when the color of the target and the background is too similar, their separability is poor. In this case, this kind of modeling method is difficult to distinguish the target and background, resulting the tracking performance of the algorithm to degradate.Harris-Mean Shift algorithm is to solve the above-mentioned problem that Mean Shift algorithm will cause tracking failure when the color of the background and the target is too similar. However, the undesirable tracking results when the target exists a large proportion of occlusion is still to be improved.Based on the above, target tracking algorithm combining Harris-Mean Shift algorithm with least squares algorithm is proposed. Describe the specific method of combining Harris-Mean Shift algorithm with least squares algorithm. Firstly, the least squares algorithm is adopted to predict the position of the target, then Harris-Mean Shift algorithm Iterates as a starting point with predicted position and obtain the real position of the target finally. This algorithm achieves acceleration by reducing the distance of vector and the convergence point when searching each frame. And tracking results is showed when the target exists a large proportion of occlusion, the results were analyzed. By comparing the experimental results of Harris-Mean Shift algorithm and improved algorithm, it is proved that the improved algorithm has stable tracking results when the target exists a large proportion of occlusion.