Abstract:

The mean shift algorithm is improved to deal with the poor performance of the current mean shift in tracking non-rigid objects with eomplex movements.First, distractions in the background are removed through background subtrac- tion based on edges. Then, GVF snake is used to extract the target contou r, which is used to improve the color histogram. In the end, mean shift is improved combining with the Kalman filter or the particle filter based on the eolor histogram. The experimental results show the real-time performance in tracking non-rigid objects and the robustness to irregular movements and he avy occlusion.

Key words: non-rigid object tracking;GVF Shake,mean shift;Kalman filter;parficle filter