Face Tracking In Video Based On Mean Shift

Face tracking is an important research subject in computer vision field, and its research results has high economic value and broad application prospects in face recognition, smart surveillance, generation computer interfaces and other many areas of civil and military, so it has been paid more and more concerns. This paper studies the face tracking algorithm based on Mean Shift to aim at the complex changes of the movements and scene in the video.In object tracking area, Mean Shift algorithm is a good tracking method, it can deal with the object matching problem between two successive frames fast and effectively. However it has the drawbacks: templates can not to be updated, a single feature is used, and the bandwidth is fixed. All these make object easily lost when the backgrounds vary, the gesture and sizes of the object change and the object is under the external disturb.This paper improves the Mean Shift algorithm by using adapted templates updated, introducing multi-features object model representation with color and texture, using ellipse fitting method to update tracking bandwidth. Viper, a performance evaluation platform for object detection and tracking, is used to evaluate the above improved methods, the results showed the new approach tracks well, and yields satisfying results even under the complicated object motion.