Data Obtaining, Organizing And Dynamic I/O For Large-Scale Scene Visualization In 3D GIS

In recent years, researches and applications related to "Digital Earth" have been given extensive attention all over the world. As one of the important supporting technologies of "Digital Earth", three-dimensional Geographic Information System (3D GIS) also developed rapidly. In this thesis, the design and realization of a 3D GIS platform--SEI3D are introduced, which are based on the research of Dalian spatial data infrastructure construction. And for the project of "3D Scene Demo System of Luan River Drainage Area", the data acquisition, organization and I/O of large-scale 3D scene visualization are mainly studied.Large-Scale 3D scene visualization is an important research subject in 3D GIS. But traditional method has a shortage on large amount of data, so only small area 3D scene could be visualized in real-time and interactive manner. There are two stages that need to be improved: one is from hard disk to main memory, in which data can be well organized and data I/O can be optimized to save time; the other is from main memory to screen, in which MRM can be adopted to improve render procedure. This thesis aims at the first stage, and following subjects are discussed.First, to discuss the acquisition of the spatial data of a large-scale 3D scene. Spatial data acquiring is an important bottleneck of geographical information application at present, but in fact there are abundant paper maps or vector maps in many cities, which should be used efficiently. In the first part of this thesis, the method of data collection for building a 3D scene based on a planar vector map is studied. To the most difficult problem in data collection: the evaluation of contour line, a suit of effective automatic assisting method is advanced, which can remarkably shorten the process period and improve the efficiency of spatial data collection. The main idea of this method is "control point congregating automatically and close control point cooperating".Second, to study the organization of the spatial data of a large-scale 3D scene. In order to satisfy the demand of data I/O in real time when scene rendered, partition and process of the spatial data according to different detail criterion should be done, and with the help of a nested index structure, the scene pyramid + sub scene regular dividing quad-tree, spatial data of a 3D scene is organized into a multi-level of detail pagoda structure. In the thesis, compatible data storage fashion and data structure for a large-scale 3D scene is discussed firstly. Then the process technique for DEM, spatial object geometry data, texture image data and scene metadata are discussed in turn. Finally, a new nested index, scene pyramid + sub scene regular dividing quad-tree is brought forward, which consumes less memory and reduces the complexity of spatial data query. The construction and application of this index are discussed too.Third, to discuss the problem of data I/O in large-scale 3D scene visualization.According to different projection fashions, the large-scale 3D scene visualization is separated into two kinds: digital sand model and real-time ramble. The elementary scheme of data I/O for each kind is discussed firstly. Because of the complexity of 3D scene visualization and the uncertainty of rambling route, the manner of real-time rambling has always been a key-point and a nodus in research of large-scale 3D scene visualization. For rambling manner, multifold optimal strategies are studied: the fundamental principle of Occlusion Culling is to reduce the I/O data amount on the basis of View Frustum Culling by taking the keep-out relation of spatial objects into account. By constructing presentations with different detail levels for spatial objects in every sub scene, Simplification of LoD increases the efficiency of data I/O at the expense of visualizing precision measurably. By utilizing the superiority of multitask operating system, the multithreading synchronous mechanism can help to implement the data I/O and scene visualization in different threads. In order...