Constructing good quality motion graphs for realistic human animation

Human animation has become an integral part of a diverse range of media, such as computer games, movie special effects and virtual training applications. Motion graphs built from motion capture data have emerged as a very promising technique for automatic synthesis of natural human motion. They use a simple graph structure to embed multiple behaviors and are well-suited for both interactive control and off-line motion synthesis. In this thesis, we address two fundamental problems with motion graphs to make them more accessible to a wide range of animation users. This thesis work especially benefits novice animation users who desire simple and fully automatic motion synthesis tools, such as motion graph-based techniques. First, achieving both good connectivity and smooth transitions in motion graphs is a difficult task. Good connectivity requires transitions between less similar poses, while good motion quality results only from transitions between very similar poses. Our Well-Connected Motion Graph addresses this problem by introducing interpolation between the user-provided motion data and minimizing the number of interpolated poses present in the graph while preserving the graph quality. Second, manually selecting a subset of motions from a large motion capture database to create a good quality motion graph is an arduous and often imperfect process. On one hand, we want this subset to be of the smallest possible size for fast motion search, and on the other hand, the subset needs to contain enough motion data to satisfy user requirements and to generate good quality motions. We cast the motion selection problem as a search for a minimum size sub-graph from a large motion graph built from the large motion capture database. While this problem is NP-hard, our efficient Iterative Sub-graph Algorithm provides a good approximation to the optimal solution and scales to large motion capture databases