Feature preserving motion compression based on hierarchical curve simplification

The authors have been studying motion database systems. When entering an example motion as the query for the similarity search of motion data, it is natural to enter it as a semantic primitive motion, i.e., "walk", "jump", "run" and so on. Mostly, one motion data consists of several primitive motions. It is necessary to divide a composite motion into primitive motions. There are no algorithms able to automatically divide a composite motion into semantic primitive motions perfectly because the semantic meanings of primitive motions are strongly depending upon the human senses. A curve simplification algorithm is used for the key-posture extraction from motion data. This helps us to divide a composite motion into its primitive motions. The key-posture extraction is also used for the motion compression. The authors propose a new efficient key-posture extraction method that hierarchically applies the curve simplification algorithm to the feature joints of a human figure model.