Motion capture data segmentation using kernel dynamic texture

Virtual human motion capture data has been widely applied in virtual reality, computer animation, etc. It is important to segment the motion data into small clips automatically in order to reuse these data. In this paper, a new method of motion capture data segmentation is proposed based on kernel dynamic texture. First we model the motion capture data using kernel dynamic texture, and then on the base of similarity measurement, we implement the segmentation method by which we can segment motion sequences into distinct behaviors automatically. Experimental results show that this approach provides the good performance.