Real-Time And Interactive Browsing Of Massive Mesh Models

3D mesh models are dominant in computer graphics. The applications employing meshes include:movies, games, computer aided design, simulation, art and history, medicine etc. A whole field called digital geometry processing based on polygonal meshes had been developed and matured and still shows vibrating vigor now.With the fast development of3D acquisition, modeling and simulation technologies, we have much more complex and accurate mesh models. Right now, mesh models of gigabytes size are not uncommon. Also, in the last several decades, the performance of CPU and GPU also improves tremendously. But, the memory bandwidth especially disk bandwidth grows much slower. So the bottleneck lies on the fact that our processor has to wait for the data stored on disk. Research on how to visualize and interact with massive models remains a hot topic in recent years. There are several famous works such as Level-of-detail (LOD) based mesh visualization system, Real-time Ray Tracing system and Far Voxels System. But each one of them fails in one way or another.In human visual system, the sensitivity to details is inversely proportional to the distance between human eyes and the detail position. At the same time, only a limited region and direction can be seen by human eyes at one moment. Those two facts can drastically reduce data amount, thus improve the performance.In this paper, we propose a novel approach for out-of-core construction and real-time interaction of massive mesh models. Our method uses face clustering on an octree grid to simplify and build a Level-of-detail (LOD) tree for the model. Each octree node leads to a local LOD tree. All the top layers of the local trees are combined together to make the basis of the global LOD tree. At run time, the LOD tree is traversed top to bottom to choose the appropriate nodes given the current viewpoint parameters. The system performance can be dramatically improved by using hierarchical culling techniques such as view-frustum culling and back-face culling. By combining a large set of technologies, our system shows good performance, better visual results and a highly scalable architecture.