Realistic human figure synthesis and animation for VR applications

Human figure animation is a widely researched area with many applications. This thesis addresses issues specific to the synthesis, animation and environmental interaction of human figures in complex virtual worlds. The techniques developed are aimed at applications such as virtual space teleconferencing, where realistic real-time animation of the human figures is desired. Visual realism of the characters dominates over the physical accuracy of the generated motion.; A layered representation of the human figure is adopted. Physical modeling and animation techniques, though effective, are too complex to achieve real time results. Implicit functions show promise in elegantly handling many issues involved in articulated figure animation that can be cumbersome for existing B-rep (boundary representation) based techniques. Real-time animation requirements, however, currently make the use of polygon-based models imperative. Approaches to improving the display efficiency of implicit functions are therefore discussed. Further, general implicit shape definitions for objects modeled using B-reps are provided. This allows an integration of B-rep and implicit function based techniques. To facilitate efficient environmental interaction, an existing collision detection and deformation model is extended and combined with constraint satisfaction for deformable objects. A realistic human figure model, allowing the application of implicit function and B-rep based animation techniques, is proposed. The implicit function muscle model and skeletal posture control a polygon based skin structure. Implicit functions detect and handle collisions in the environment homogeneously and efficiently. Precise collision contact surfaces are generated and physical characteristics of muscles in systems employing force feedback are simulated. Clothes are modeled as textures mapped on a geometric skin. Characteristic wrinkled textures for garments are synthesized and wrinkle formation is carried out by texture morphing and wireframe displacement controlled by the posture of the animated human. The effectiveness of the developed concepts and proposed models are illustrated by their implementation within a virtual space teleconferencing system.; This thesis contributes to object modeling and animation by successfully marrying implicit function and B-rep based methodologies. The human figure models developed are elegant, efficient and showcase this hybrid approach. These solutions can be easily applied to existing real-time human figure animation applications where visual realism is more important than an anatomically and physically precise model.