Splicing partial body motion video sequences for motion synthesis

Motion synthesis has become an important research area in computer animation. When synthesizing new motions, recent partial body parts motion synthesis methods allow two upper body motions to be swapped if they have similar lower body motion. Although such requirement can ensure natural motion, it limits the possibility of new motions that can be synthesized using this method. In this paper, we propose to relax this requirement by considering also lower body motions that have only similar functionality and regularity. To facilitate the similarity measurement, a new time alignment algorithm is proposed so that motion sequences which have similar regularity can be easily identified. Our results show that the proposed algorithm can significantly improve the reusability of motion data for motion synthesis.