Skip to Main Content
In spatial face detection stage, in order to eliminate the influence of luminance, a two-dimension Gaussian distribution function, based on the chrominance plane of YCbCr color space, is constructed, then the moving skin area is determined by the luminance component Y, the moving face is detected by support vector machine classifier. In face temporal tracking stage, the eigeface similarity measurement of consecutive frame is got by Bhattacharyya coefficient, which can give us the trace of human face. In order to make the tracking scheme robust, the Kalman filter is used for updating face model, the mutual feedback scheme benefits both face spatial detection and temporal tracking. In order to deal with face occlusions and whether or not to update face model, we take advantage of hypothesis test. Finally, the comparison experiments between the proposed with other methods are given. Numerical simulations show the proposed algorithm can accurately detect and track human face.