Constructing And Rendering Of Multiresolution Representation For Massive Meshes

With the advance of 3D scanning, computer-aided design and scientific simulation technologies and with the push by application needs, massive meshes containing over billions of geometric primitives are becoming popular. It is difficult to render these massive meshes interactively. The out-of-core multiresolution technique, which is one of the most efficient approaches to improve the rendering performance, has become a hot topic in the field of computer graphics. However, there are many difficulties in the design and implementation of this technique for massive meshes. Firstly, as geometric data and auxiliary data structures of massive meshes can not be completely loaded into memory due to limited memory size, the traditional multiresolution modeling and rendering algorithm can not be directly applied to the massive meshes. Secondly, with limited bus bandwidth and CPU processing power, the construction process of multiresolution representation for massive meshes often takes much time, which is not conducive to system debugging and real-time applications. Based on intensive study on massive mesh simplification and multiresolution techniques in recent years as well as analysis and comparison of key ideas of various methods, this dissertation realizes the parallel construction of multiresolution representation for massive meshes. The parallel construction method improves the construction speed effectively. Subsequently an algorithm for constructing and rendering out-of-core multiresolution representation for massive meshes is proposed. This algorithm takes account of the speed and rendering quality. Then an algorithm for constructing and rendering out-of-core multiresolution representation for massive meshes with complex topology and large local depth differences is proposed and realized. Lastly the above parallel construction algorithms are extended to Grid envirionment.The contributions and novelties of this dissertation are as follows:1. Based on the external memory octree construction algorithm of multiresolution representation for massive meshes, this dissertation proposes a sub-tree task partition mechanism and a benchmark measuring based dynamic construction task management mechanism, and realizes a parallel construction of multiresolution representation for massive meshes with the load balancing. This parallel algorithm improves the construction speed effectively. 2. A novel approach for constructing and rendering out-of-core multiresolution representation for massive meshes base on adaptive spatial clustering is proposed. This approach not only improves the rendering quality, but also maintains the efficiency of spatial clustering algorithm. It keeps the details of the original model by using adaptive octree partitioning the mesh bounding box, and reduces the average cache miss rate through mesh layout optimization for triangles contained in the octree node. At rendering time, the multiresolution representation structure is refined and rendered with view-dependent LOD selection. At last, the method employs data prefetching strategy to hide I/O latency and then improves the rendering performance further.3. A novel approach for constructing and rendering out-of-core multiresolution representation for massive meshes base on surface segmentation is presented. This approach provides vertex-grained local refinement and generates the optimal rendering quality. This algorithm represents the model as a double hierarchy: a clustered hierarchy of progressive meshes for coarse-grained selective refinement and a vertex hierarchy of progressive meshes for fine-grained local refinement. In order to improve the speed of local refinement, a parallel view-dependent refinement mechanism is realized by using GPU, which also reduces CPU load and makes it available for data prefetching, then hides I/O latency further.4. Based on the previous work of interactive grid visualization system(GVis), this dissertation extends its runtime environment layer to support the parallel construction of multiresolution representation for massive meshes, consequently realizes dynamic resource allocation and scheduling as well as enhances flexibility, fault tolerance and scalability of parallel construction process. Additionally, this dissertation extends its grid portal to provide display service of massive dataset to end users.