Study On Virtual Human In Virtual Environments

This dissertation performs some efforts on improving the existed modeling, animation and rendering techniques of virtual human. The main contributions of this dissertation include:In the area of avatar modeling, we enhance the way to keep adjoining segments stay connected in model customizing algorithms, which typically modify the shape of a generic model with shape features extracted from pictures, and generate a model which both facilitates the handling of animation and reflects shape details in that particular person. These algorithms have low hardware/software requirements but can easily cause initially connected segments to disconnect. We prepose to parameterize the points in a segment in a way featuring axes and cutting lines, and ajust points near the joint with smoothly varying local modification, and thus maintain smoothness between adjoining segemnts. Other methods maneged to achive similar level of smoothness by imposing strict requirements on the topology of generic model, but our's does not need any such restriction.In the area of motion control, we propose a real time path synthesis algorithm, which searches suitable frames in captured human locomotion data according to a user specified track and regroup them to form a new animation sequence walking on that track. The core of the algorithm is to cut, match and regroup captured motion data with the help of motion graph. But existing methods use slow algorithms to construct and search the motion graph, and thus can not meet the real time requirement. We propose to reduce the number of jumping nodes by clustering similar frames when creating the graph and to avoid redundant computation by taging and caching when searching the graph. Experiment results show that our algorithm can fulfill the real time requirement.We also propose an adaptive algorithm to render large-scale virtual-human-including dynamic scene in realtime. By integrating the techniques of visibility culling, level of detail and image caching, and by choosing a rendering schemes for every object based its importance and the overall user-specified time budget, the algorithm can render the scene to acceptable quality in realtime. The scheduling algorithm compoutes importance of dynamic object based on its velocity and computes importance of virtual human based on view direction and crowd density, thus making it more suitable to handle virtual-human-including scenes.Last, we introduce an application systems developed with avatar techniques. It is a virtual rehearsal simulation system for group calisthenics, which represents each participant with virtual human, provides path-planing functionalities and renders the result in real time for rehearsal.