This paper proposes a novel template adaptive updating strategy as part of Mean Shift framework which is based on feature matching filter in the scenario of head tracking. In accordance with the color space statistical information of tracked objects in different states, by using the color space kernel density estima- tion matching filter, this method decides on the specific pixels that need to be updated in template. The experimenttal results show that the algorithm is an effective solution to the rotating head and object block which leads to the mismatch of template. It also achieves the continuous tracking of the head, and has very good tracking results. The overall stability and robustness of the tracking algorithm are improved.