Automated face tracking with self correction capability

摘要

Face Tracking is one of the most challenging topics in computer vision. Various face tracking methods have been proposed. However most of them have not ability to correct error and divergence in face tracking process. In this paper we propose a new method for face tracking using face detection and object tracking simultaneously to utilize their advantages at once. For minimizing error and divergence from target, we propose a feedback system based on Local Binary Pattern (LBP) and several rules to provide this opportunity that detection and tracking systems can cooperate with each other, so that ability of one system cover disability of another one. We demonstrate the performance and effectiveness of the proposed method on a number of challenging videos. These test video sequences show that proposed method is robust to pose variations, illumination changes and occlusions. Quantitatively, proposed method achieves the average root mean square error at 6.78 on the well-known Dudek video sequence. Experimental results show reliability of the proposed method.