| Animated, human-like characters play an important role in virtual reality, interactive entertainment and simulations. The human body is a very complicated system, and thus creating realistic-looking human motion for animated characters is a difficult task. Synthesizing arbitrary motion for a humanoid character is a difficult control problem. Motion capture data is readily available and can playback an actors movements on an animated character. However, it is impractical to capture all variations in motion in order to accommodate the range of activities of an animated character in the virtual world. As such, animators have utilized motion editing as a means to modify motion capture data to suit the needs of the virtual environment. Generally speaking, motion editing is a difficult task and the more the motion capture data is modified, the less realistic the resulting animation. This thesis presents three novel approaches for motion editing that produce results that were either difficult or impossible to produce using traditional motion editing methods such as keyframing, while still maintaining realistic looking animations. These tasks include: (1) Editing the style of a motion from one type to another while maintaining the motion semantics, (2) editing motion capture data to accommodate moving characters, moving obstacles and moving reach targets, (3) editing motion capture data in order to accommodate large disturbances from other characters and objects in the environment. In addition, this thesis presents a framework and software system for editing and synthesizing character animations. |
