Research On Surface Reconstruction Of Spatial Objects From Point Cloud And Its Multi-Resolution Represention

To describe spatial objects and its related information comprehensively and accurately is a key research of digital city. Resorting to its high speed, high efficiency and high accuracy, 3d laser scanning technique provides dependable guarantee of data for successful building of digital city by describing spatial objects with dense point clouds. However, for lack of necessary topology information among sample points, applications such as spatial analysis of G1S, collision detection of virtual reality and virtual cutting of geological 3d GIS can not be realized based on point clouds. Besides, as a data acquisition technique which holds high efficienty and high accuracy simultaneously, 3d laser scanning technique's demand of computer resources and storage spaces is high.To fully bring 3d laser scanning technique into playing roles in building of digital city, and to alleviate computing resource's burden from super-large data, post-processing of point cloud has been studied in detail. As is known, when 3d laser scanning techinique is applied in building of digital city, density of sample points often is not uniform, blind sample areas often occur, and point cloud is super-large et al. In such cases, to correctly establish topological relationship among sample points is the aim of this dissertation. Above all, multi-resolution representation of 3d surface models and encoding 3d surface models with triangle strips are also studied. Achievements are as follows.(1) A Delaunay-based surface reconstruction approach for unorganized sample points has been designed and realized. By projecting sample points onto local fitted tangent planes, constrained Delaunay triangulation is applied to the projected points. And the connection of these points is mapped directly back to 3-dimensional space. Then we obtained topological relationship among these points in 3-dimensional space. By lower dimension of sample points, the presented approach can deal with super-large point cloud with ease. Furthermore, the presented approach can detect areas of low sampling density and boundaries of sampled surfaces, and so it can be used to reconstruct surfaces with or without boundaries.(2) A feature-constrained surface reconstruction approach for point cloud is presented. Firstly features are extracted from point cloud under rules of curvature extremes and minimum spanning tree. By projecting local sample points onto fitted tangent planes, extracted features are used to guide and constrain the process of local triangulation and surface propagation, and so topological relationship among sample points can be achieved with reason. With constructed models, a process named consistent normal adjustment and regularization is adopted to adjust normal of each face, and so correct surface models are achieved. Experiments show that the presented approach inherits the convenient implementation and high efficiency of traditional incremental surface reconstruction method. Meanwhile, it avoids improper propagation of normal across sharp edges, and so applicability of incremental surface reconstruction is greatly improved. Above all, appropriate k-neighborhood can help to recognize un-sufficient sampled areas and boundary areas, and so the presented approach can be used to reconstruct both open and close surfaces without additional interference.(3) An algorithm for constructing LOD models using the rule of quadric error norm and edge collapse opeartor has been presented. Recurring to quadric error norm, distortion of models can be measured effectively after each edge collapse operation, which is the basis for arranging all candidate edges. Such action is to ensure that distortion of each edge collapse operation is minimal. In the process of generateing LOD models, half space testing is introduced to judge the validity of each edge collapse operation, which is to avoid self-interaction of generated LOD models. Most important of all, both triangular faces at the boundary and in the interior of the surface model are treated impartially using our new algorithm, and so ability to approximate original high resolution models is improved, quality of the result LOD models is effectively guaranteed.(4) An adaptive, view-dependent and topology-preserving mesh simplification approach is presented. Progressive mesh is used to represent original high-resolution models, and hierarchical binary trees are constructed based on the turn of edge collapse operations. In the process of real-time creation of view-dependent LOD models, several parameters related to visual effects are considered, they're view point, view direction, field of view, normal of surface, and incidence of a certain vertex et al. In contrast, the presented approach can be used to selectively refine or coarsen a certain area of a model. As a result, several different levels of detail may co-exist across different regions of a certain result model. Under the precondition of ensuring rendering effects, the presented approach greatly diminishes the amount of data to approximate original high-resolution models, and so rendering speed of result models is rapidly accelerated.(5) A graph-based approach for efficient generation of triangle strips from 3d surface models is presented. Recurring to support of graph-theory and topological relationship among triangle faces, the approach generates triangle strips under the condition of a global criterion. Moreover, it will join any two triangle stips whenever possible. By separating the two procedures named route-extracting and encoding, the new approach has the ability to keep normal of all triangles consistent automatically throughout the whole encoding process. In constrast to available methods, repetitive normal adjusting operations are avoided. Additionally, recurring to the correspondence between dual graph and original triangle mesh, a triangle mesh based encoding scheme is adopted, which ensures the approach's high efficiency and high quality. Experiments show that our approach is easy to understand and to implement. Most important of all, with the increasing complexity of surface models, our approach exhibits unexampled advantages.Based on above achievements, prototype system serves for post-processing of point cloud is designed and developed, and it integrates with RiSCAN Pro weamlessly. The prototype system not only provides effective experimental environment for our research, it also provides basis for 3d modeling from point cloud data acquired by 3d laser scanning technique.