Research Of Real-time Rendering Based On Solid Voxelization

Real-time rendering is the core part of scene simulation and game. Triangle meshes are used for most of the scene representation. But triangle meshes will cause aliasing problems due to fixed sampling rate. Voxelization methods and volume rendering methods based on it have been actively researched recent years, and all new generation engines recently launched have volume rendering methods within them. But in China, researches related to these methods are rare, so in this dissertation research has been done around the voxelization method, and also the whole rendering pipeline has been explored. Especially, the filling error problem for scenes with interfered models is solved, and the sampling and transparency reconstruction problem in approximate voxel cone tracing is analyzed.The works of this dissertation mainly contained are:1) A new pairing method used in solid voxelization process which handles interfered models correctly is proposed. The method is implemented as a parallel algorithm on GPU, which utilize the property that geometry fragments in any line are paired. The experiment results show the correctness and efficiency of the method.2) The approximate voxel cone tracing method on uniform voxel volume is implemented and the rendering result influenced by different sampling rate is compared. A transparency reconstruction approach using optical depth based on cone tracing method is also proposed. The experiment results show the pros and cons of different sampling rate and transparency reconstruction methods.3) To facilitate the research, a renderer which using modern graphics hardware APIs has been developed. The renderer provides the framework, utilities and abstraction of the graphics API. All the algorithms in this dissertation and a scene roaming project are powered by this renderer and have got good results.