A multiple faces tracking system was presentedbased on Relevance Vector Machine (RVM) and Boostinglearning. At the first frame, a face detector based onAdaBoost is used to detect faces, and the face motion modelsand face color models are created. The face motion modelconsists of a set of RVMs that learn the relationship betweenthe motion of the face and its appearance in the image, andthe face color model is the 2D histogram of the face region inCrCb color space. In the tracking process, differenttracking methods are used according to different states ofthe faces and the states are changed according to thetracking results. When the full image search condition issatisfied, a full image search is started in order to find newcoming faces and former occluded faces.
展开▼