Research Of Real-time Rendering And Model Simplification Algorithm About Large Scale Terrain Environment

With the recent explosive growth of geography information system and aerospace technology, especially in the last few years the digital Earth develop rapidly, people less and less satisfy with the usage of traditional 2d (two dimension) techniques for expressing the geography information. How to deal with a mass of terrain data that obtained from the satellite or airship, how to visualize them and generate 3d (three dimension) scene, and how to gain the real-time alternant function for consumer, are hot issues in the computer graphics research field.Three dimension terrain visualization is a subject that study how to display,simplify and imitate the digital terrain model or digital elevation model, which belongs to a branch that learn of calculator graphics. 3d terrain visualization is divided into three steps such as the terrain generation, the terrain simplification and the terrain rendering. Digital Elevation Model(DEM) is always used to express the terrain, whose data structure is mesh data structure(grid data structure) with the characteristic of huge data size and redundancy. While the accuracy of these 3D models grows with the development of the 3D scanning and modeling techniques, the volume of their data sets grows even rapidly. That brings great pressure to the rendering, transmitting and editing system of computers, although there is a great development of the computer's operation speed, memory capacity and graphics accelerator. One way to solve this problem is to simplify these complex 3D models and to perform multi-resolution modeling.The key technology in this paper is LOD(Level of Detail) algorithm. The main idea of LOD is: If the object of the hierarchical structure assembles and describes a scene by having it more,That is to say that the object in the scene has a lot of models, the difference among its model lies in the description degree of the detail, When showing in real time so, less detail model can be used for raising the speed of showing. While showing in real time,the choice of the model depends on the important degree of the object, The important degree of the object is accounted for many kinds of factors, such as area on the vision space. LOD technology is developed so far, has experienced dispersed LOD, continuous LOD and multi-resolution LOD three stages.Digital terrain model has natural consistency with the structure of tree, so quad-tree is usually applied to manage terrain data. Multi-resolution LOD model on the basis of quad-tree is the focus of present research. On the basis of an error threshold value of the setting, a joint evaluation formula is brought forward to select the terrain node. Through the formula, many useless nodes can be eliminated so that the speed of rendering will be improved.Though multi-resolution LOD model uses a better kind of model of simplifying efficiently, but still there are some difficulty parts that need improving badly. Among them, a dose of joints coming from mesh of different resolution ratios will produce the crack. It is one of the subjects that a lot of people are studying how to design a kind of elimination crack algorithm trying to be high-efficiency on the basis of guaranteeing to enhance the result.In this paper, after much of systemic study and analysis of three dimensional terrain simplification and visualization, deeply investigates the theory and algorithm of three dimensional terrain model simplification and real-time imitate technique. This paper's main contribution include: (1) Studying and analyzing a lot of different format DEM data files, This paper presents a data file format which can save memory and read/write rapidly, and implements the transform of many kinds of formats. So it is not only saving the memory and increasing the speed of reading/writing data, but also strengthening the three dimension terrain visualization system's capability of distinguishing the different format DEM data.(2) Because of the huge data size, it will influence the system efficiency that changing data between the high-speed internal memory and the low-speed external memory. This paper uses Lindstrom's method: implement out-of-core data load by using"MapViewOfFile"function.(3) During the recursion subdivision of the quad-tree, the main idea of the node error evaluate method is based on the difference of the node elevation with the characteristic of easily and rapidly calculating. This method is fit to be a standard method of model simplification algorithm. But there are two problems: one is during the simplification lots of details are losing, another is the redundancy. In order to solving above two problems, this paper presents a new model simplification algorithm based on the divergence function of the simplification area's normal vector. This algorithm evaluates the relationship between the triangles with the highest resolution and current triangle before subdividing it. This algorithm's dominant is considering both local details and global terrain features.(4) Owing to the cracks exist between the adjacent nodes with the different resolution, A typical restricted quad-tree method is proposed in, this method generally subdivides all the nodes with low resolution. For each node in the restricted quad-tree, a restricted relationship with its adjacent nodes is created beforehand. Thereby, the adjacent nodes that satisfy the restricted relationship is also subdivided when the current node is subdivided. As a result, many redundant triangles will be produced. In fact, the repair operation of cracks should be performed between all adjacent leaf nodes. However, the cracks are not obvious even invisible in some flat terrain. Accordingly, this paper evaluates the terrain features before repairing the possible cracks that exits between the adjacent nodes with the different resolution. Differing from the height difference of the vertex, this evaluation function is based on the normal vector angle of the adjacent nodes. Owing to this method takes the surface information into account, we can just only repair the real cracks and not produce redundant triangles.In order to illustrate the validity and the efficiency of this paper's method, we conducted several experiments on real world data. The platform of the experiments is a PC with a Pentium4 2.0GHz CPU and 512MB main memory. The algorithms are coded in VC++. In experiments, As a result, this method is relatively efficient and there are not redundant triangles in our seamless model.