Researches On Object Tracking In Video Sequences Using Mean Shift Algorithm

Detection and tracking of moving objects are key to digital video analysis. It has wide applications in smart surveillance systems, next generation computer interfaces, and many other fields in need of such automation.This paper actualizes a system capable of tracking and detecting objects under the DirectShow environment, and concentrates on Mean Shift algorithm. Mean Shift algorithm is a good method for objects tracking. It deals with the object matching problem between two successive frames which is capable of keeping the objects being tracked in the center of the frame. And it runs both fast and effective.However templates need to be updated, if not, objects being tracked are easily lost in circumstances where a clutter is present or when an object moves fast. This paper improves the Mean Shift algorithm by introducing adapted templates updates, using redefined kernel bandwidth to re-locate positions of objects. Then it combines the Mean Shift with Kalman filter, the later is used for estimation, the Mean Shift then computes the optimal location for objects being tracked. This new approach tracks well according to our experiments, it yields satisfying results even under camera motion.