Synthesizing Human Motions: Various Data-Driven Approaches
By now, the recording of motion data is a standard technique yet of increasing importance. Accordingly, the work at hand dicusses a variety of methods for searching data bases, semantic annotation and synthesis. Most of the available motion capture data have been used for one specific project only and never reused although all these motion data provide valuable insight on human motions. Investigating the benefits of having all this data, possible fields and natures of arising applications works especially towards achieving a higher level of reusability. Basic techniques designed to handle a large amount of motion capture data such as methods for fast similarity search and for automatic annotation of motion capture data are presented. Three different methods of motion synthesis emerge: tensor based multilinear representations constructed from annotated motion sequences is one, while enhancing given motion sequences using a technique for motion texturing is another. Finally, a technique for motion synthesis from sparse key frames also employing the search algorithm is introduced.