Abstract:

A fast face capturing and tracking method in motion image is proposed, first the face area is located according to the template matching filtering, next the SVM serves as a classifier for final confirmation, then a CamShift tracking method based on MeanShift is used to keep the tracking, and parameter identifying and state estimating are carried out through Kalman filter while occlusions take place. The method has been demonstrated to be real time, efficient and robust by experiments.

Key words: face detection, face tracking, SVM, MeanShift, CamShift, Kalman filter