Research On Human Face Tracking Algorithm On Video Image

Face tracking plays an important role in the computer vision field.How to track the target face dynamically and accurately has become a valuable reserch subject.According to the different background of the vedio,face tracking is classified into two kinds in this paper.One kind is that the backgroud keeps static,and the other is that the camera moves as the target's movement.When the background of the vedio is static,in consideration of the fast moving of the target face,Camshift algorithm may generate hysteresis phenomenon because of its iterative calculations.In order to solve this problem,this paper made use of the inter-difference method to detect the moving objects,and then used YCbCr complexion model to obtain the skin color regions.Finally,it gained the target face by limiting the size and the aspect ratio of the skin color region.It is proved that this method's running time is short and it is possible to track human faces real-timely and accurately.In view of the dynamic tracking background,this paper combined Camshift algorithm and facial feature points locating algorithm.Firstly,it adopted ASM algorithm to locate the two medial canthus.And then it calculated the size and deflection angle of tracking window depended on the distance of the feature points.Experimental results showed that this algorithm can overcome the defect of traditional Camshift algorithm,which is often disturbed by the object whose color is similar to the target's color.To some extent,this algorithm enhanced the anti-jamming capability,and completed automatic face tracking in the video.In order to prove the validity of the two methods,this paper respectively used the two methods to conduct experiments.The experiment results proved that both of the two methods can tracking the terget face accurately and dynamically in their own application scenarios.