Research On Huge Fine Spatial Data Management

Grid point cloud/depth image from terrestrial LIDAR, point cloud model and partial final products from data processing are chosen as the research objects of this paper and series of methods on data organization and management based on grid point cloud are put forward, such as:design and construct of QMBB tree spatial index、 segmentation and reconstruction of grid point cloud based on QMBB tree, automatic and semiautomatic regular geometrical model extraction, hybrid spatial index of3D k-d tree and Octree method for point cloud model, data organization and management methods for huge data set of point cloud, organization and visualization of spatial data model etc. The Minimum Bounding Box (MBB) of all the objects in the same coordinate is computed and R-Tree spatial database index is built according to their MBB. The technology of hybrid spatial index mentioned above and level of detail is adopted when the object is too huge. A set of system including3D spatial index、point cloud organization and storage and visualization of huge data is formed. Research works are concluded as follow:(1) The conception and features of grid point cloud is introduced in detail from the original point cloud. Huge spatial data processing, storage and management are put forward from grid point cloud, including2-d and3-d hybrid spatial index, multi-scale level of detail of segmented data, extraction of regular geometric model, construction and organization of CSG model, point cloud model management and3-d spatial data storage and visualization etc.(2) Quad-MBB Tree index with the character of2D and3D is raised. Concept of Quad-MBB tree is given. A flexible hybrid spatial data segmentation and index method for the original point cloud or depth image are put forward according to the unique, raster and huge of grid point cloud, which combines features of Quad-Tree index and MBBs of objects. The Quad-Tree and3D MBB tree spatial index are built simultaneously for grid point cloud model in the order from top to bottom and from bottom to top.3-d R tree is adopted to manage the multi-scans data by its root MBB. The serial spatial index and the compression data structure are designed and two organized kinds of data with other attribute data are stored in the large commercial database. (3) An improved algorithm to extract feature geometric model from3d grid point cloud is study and designed. An optimized algorithm combined conventional RANSAC method is designed to extract regular geometric model automatically from grid point cloud without the need of normal and curvature computation. The geometrical parameters are got by iterative computing through seed point selected with RANSAC algorithm and adaptive parameters determination and then the feature point is clustered by the predefined threshold value of distance from point to geometrical model. The total geometrical model in the one sight is got by fitting the feature point in QMBB’s leaf node and merged among different leaf nodes and then the TIN model is transformed from the total geometrical model for the prepare work for building CSG model.(4) The generation methods and characteristics of the huge point cloud model are analyzed and summed up on description of point cloud model concept. The hybrid spatial index based on Octree and3D K-D Tree is designed to deal with point cloud model. The fast data retrieval and real time visualization of large scale point cloud model is realized through hybrid spatial index combined with the technology of window cutting and multi-resolution level of detail. In order to make up the spatial database’s drawback on large scale spatial data model visualization,3d R-tree index is designed to deal with spatial data model based its Minimum Bounding Box.With the theory and research production above, a system is developed with c#and c++program language on the platform of Microsoft Visual Studio.NET. Database of Oraclellg is applied, and the PL/SQL program language is used to design the3D object data model. ODP.NET is applied to realize I/O operations of the database. The prototype system is implemented with OpenGL graphic library. Finally, the terrestrial point cloud of the Forbidden City, Beihai Park, and National Stadium are used to validate the feasibility and the efficiency of the methods above.