Research Of Human Motion Editing And Synthesis Technology Based On Motion Capture Data

With the advancement and popularity of the motion capture technology, acquiring the human motion data efficiently and quickly has become a reality. Motion capture data has been successfully applied to computer animation, video games, television stunt, etc. However, as the data is captured with specific people and specific scene, it causes great inconvenience to reuse the data itself. As a result, the research on how to edit, manipulate and reuse the existing data becomes a hotspot in the motion capture field.Reuse of the motion capture data mainly includes two aspects which are motion editing and motion synthesis. Motion editing researches how to edit a piece of given motion data so that it can be used for new character, new scene or new constraint. Yet motion synthesis studies on the process of dealing with some pieces of the motion data to produce a piece of new motion. The thesis researches the theory and key technologies in the motion editing and motion synthesis fields. The work mainly focuses on motion segmentation, keyframe extraction, motion style translation and motion graph.1. The quaternion distance is implemented to denote the difference of the human pose. To alleviate the foot sliding problem in the motion reconstruction through the interpolation between keyframes, the Inverse Kinematics based on the pseudo Jacobian solution is introduced, and more realistic reconstruction effects are obtained in our experiments.2. The PCA-based motion segmentation algorithm is implemented to automatically segment a long motion sequence into distinct behaviors.3. A method of time warping is presented which is based on low-dimensional curve analysis in the subspace. The mean and variance in the quaternion space are used to character the motion style. The smooth style translation from one motion to another is realized.4. To construct a motion graph, quaternion distance is introduced in finding the transition points, and the Dijkstra algorithm is implemented for searching the nearest path between two edges in the motion graph. Through creating transition movements, we can synthesize a piece of smooth motion.