The Study Of Dynamic Target Tracking Algorithm Based On Mean Shift

Dynamic target tracking is one of the most important content in the machine visionfield. its application is extremely wide, such as transportation, medical, military, andhuman-computer interaction. Most tracking algorithms are proposed for the problem ofone particular aspect. The mean shift tracking algorithm is one of commonly used dynamictarget tracking algorithm. it uses color histogram to describe the target during the targettracking process, so its robustness and real-time performance are very good in the relativelysimple background environment. It also has a good tracking effect on the Morph target. Butthe mean shift tracking algorithm has some defects, according to this, this paper proposesthe following improvements:Since the mean shift tracking algorithm can not effectively track the targets whichmove too fast or occluded heavily, we proposed the tracking algorithm which combines themean shift algorithm and Kalmanman filter. The basic idea of the improved algorithm is:when the target moves too fast, Kalman filter is used to predict the position of target in thecurrent frame, and mean shift algorithm iteratives based on this position, so that it not onlyreduces the number of iterations, but also the target will not be lost; When the target isoccluded, Kalman filter can predict the position of the target in the current frame, and thenmakes this position as the true position of the target in the current frame, until the distancebetween position predicted by the Kalman filter and optimal position converged is less thanthe threshold value, which represents that the target is not being occluded, Then we cantrack the target based on the mean shift tracking algorithm. Through two groups of contrastexperiments, the experiments result shows that the improved algorithm can not only trackthe moving target effectively, but also the real time is quite good.For the problem that the kernel bandwidth of the mean shift algorithm remainsunchanged all the time. We proposed the corresponding improved algorithm. The basic ideaof the improved algorithm is: since the target becomes smaller gradually, and the trackingwindow width becomes lager relatively and there is much more background in the trackingframe which is not helpful for tracking, we proposed the method of plus or minus tenpercent bandwidth, which makes the size of the window width change as the size of the target changes; when the target becomes lager gradually, and the tracking window widthbecomes smaller relatively, which leads to the tracking window convergence in the localarea, the centroid is positioned based on the mean shift algorithm, and the according to thefeature points matching, the tracking window is updated. The contrast experiment showsthat the improved algorithm can not only effectively track the moving target, but also thetracking window can change as the target changes...