Content-Based Human Motion Retrieval And Motion Synthesis

It is an important research topic in the areas of computer animation, virtual realism and human interaction to create realistic and natural human motions for virtual human characters. Due to its high fidelity of motion capture data; recently motion capture based animation techniques have become an interesting research issue in the field of Computer Graphics.With the rapid increase of motion data, example-based motion synthesis technologies are the promising research field in human motion animation, where motion database acts as the key role. This causes a problem: how to find desirable information in the"ocean of tremendous motion data"for animators. This dissertation commits itself to researches on key techniques of motion indexing and retrieval. Our research work will help users to find desired motion data from motion database, which can serve as successful motion synthesis materials.Main innovations of this dissertation consist of four aspects as follows:1. To present a motion normalization algorithm that enables all skeletons in database to share the same skeleton lengths. Usually, motions in database are obtained from different source. As a result, skeleton lengths are probably different, which will influence the similarity comparison between motion clips. Our algorithm simplifies constraint conditions of existing motion retargeting algorithm so that such preprocessing work can be processed with an on-line way.2. To introduce Muller et al.'s 31 geometric features (in SIGGRAPH 2005) into our clustering algorithm, which can significantly reduce the comparison times between motion frames during the construction of motion graph. In addition, clustering scheme helps to solve the problem that automatically selects suitable thresholds for different motion types. Because of the high dimensions in motion data, the"curse of dimensionality"will hamper the effectiveness of current clustering algorithms. Such geometric features are invariant under global position...