Face Tracking Based on Skin Color and Motion Prediction

Aimed at the disturbance caused by illumination variations and similar skin color regions in the video scene which will greatly affect face tracking accuracy, and in order to enhance the real time performance, a novel approach for face tracking based on skin-color information and motion prediction is put forward in this paper. The major idea is to take the Gaussian skin-color model based on YCrCb space as face location model, and in order to increase the real-time performance of skin-color detection, GM(1,1) model is combined to carry out face prediction so as to reduce searching areas. By exploiting least-square method to determine constant c in time response function of the GM(1,1) whitenization function, the prediction accuracy can be greatly advanced, satisfying the demand for prediction model owing to the randomization of human face movement. Experimental results show that the proposed method has such advantages as fast tracking speed and good robustness. Keywords-Skin-color Information, Motion Prediction, Face Tracking, GM(1,1) Model