Surfel-Based Geometric Model Representation And Multiresolution Rendering

With developments in 3D scanning and modeling technologies and improvementsin computer hardware conditions, 3D models are introduced into a variety ofapplications, such as geologic study, protection of cultural and historyrelics, games andamusements etc. Modeling and rendering of 3D models is a technical base and a keytache in those applications. Owing to increasing complexity of 3D geometrical modelsand growingdemands for advanced data processing, point-based geometryhas gained awide attention. The best advantage of point-sampled objects is that it do neither have tostore nor to maintain globally consistent topological information. Since compared withtraditional polygonal meshes, point-sampled objects will bring a high flexibility ingeometryprocessing, especiallyinrendering ofhighlycomplex 3Dmodels. However, apoint-sampled object has a large quantity of data, composed of millions of sampledpoints, even more in general. So how to store and transfer these data efficientlyis a keyproblem to be solved. Studys in this aspect have concentration on point-based meshesreconstruction and point-sampled modeling with a great efficiency.With KD-tree as data structure in our research,anisotropic quantization is proposedbased on the property of equal space partitioning of the midpoint-split KD-tree. Thismethod is extended to re-quantize the coordinates of the sample point position indifferent quantization precision. The quantization result is represented as encode of thespace partition manner and partition result during the KD-tree construction. Theanisotropic quantization not only reduces the volume of geometric data contained in apoint sampled model, but also decreases the quantity of sample points while thenumerical precision remains constant. As a result, the space consumption is reducedfurther. The sequential storing of KD-tree and the building of multiresolution storingstructure and the multiresolution rendering are implemented too by bread-first ordertraversal of this KD-tree. In addition, the hybrid point and polygon rendering isimplemented and surmounts the obstacle to large flat surface representation of pointsampleset.