Research On Object Detection And Tracking In Video Sequences

The detection and tracking algorithm of motion objects is one of hot spots in the field of computer vision. It’s also the key of video surveillance system. A growing number of algorithms have sprung up. According to the difference of the target tracking and scene, tracking algorithm is not the same taken.In terms of motion objects detection, three common approaches in the field of motion object detecting are analyzed including background subtraction, temporal differencing and optical flow. In addition, the principle of the algorithm is analyzed in detail, and contrasted the advantages, disadvantages and applicability of them. For the above algorithm, analyzes the principle of three-frame-differencing. This algorithm reduces the target detection in "empty" and "double". Also it improves the accuracy.In terms of target tracking, fast moving and shade have always been a hard to deal with. Only deal with the shade and fast moving object tracking in video, the accuracy of the algorithm to improve, robustness and accuracy can be more perfect. On the basis of understanding the theory of Mean Shift algorithm, adaptive tracking box shows the algorithm is applied to the simulation results. The experimental results show it applied to the fast moving target and severely block movement target tracking failure.The Mean Shift algorithm is combined with kalman filter of a tracking algorithm. However, the large proportion of shade has been remained. Then on the basis of the fusion algorithm Bhattacharyya coefficient to judge whether the introduction of the similarity function appeared seriously blocked situation. Finally, the simulation experiment results show that the algorithm of video moving object is severe occlusion of relatively stable.