A distributed hybrid rendering algorithm for highly complex scenes

A distributed virtual reality system based on a hybrid rendering scheme is presented in this paper. Unlike the traditional image based rending algorithm, our approach divides the local scene within the frustum of the current viewpoint into two layers, i.e., the front layer and the back layer. Objects in the front layer are rendered in real time with a geometry based hardware z-buffer algorithm. To generate the image of back layer, we first perform a nonlinear transformation to all visible points of a keyframe adjacent to the current viewpoint to reveal their coplanar characteristics. A hierarchical quadrangle approximation of the local scene is then set up, the intermediate image of the back layer at the current viewpoint is then obtained by re-projecting these quadrangles onto the current image plane, gaps that might appear on the intermediate image am filled by bi-directional backward warping. As our approach consists of two independent phases, it can be easily implemented in parallel or distribution mode. Experimental results show that our approach has a satisfactory performance