Point-Based Model Reconstruction And Rendering Technology

Recently, as more and more highly detailed geometric models are required in 3D geometric modeling and many graphics applications, the scale and complexity of the model data have dramatically increased. If the traditional triangle-based rendering methods are used, then we should first generate the triangular plates according to the data cloud points, that's a time-cost and error-prone job. If a method which can be used to directly molding and rending the object's surface from the sampling points, it will be of great practice significance.Under the application and requirement, a point-based molding and rending method is developed. Compared with the traditional triangle-based method, the point-based molding and rending method have many advantages. For instance, it needn't record the connection relationship among points, when rending it also needn't maintain the topological consistency of the model. So resample job can be done very easy, and a flexible hierarchical structure can be build to satisfy multire-solution rendering. Except that the point-based molding and rending method is very suitable to rending irregular models like feature, hair, flame, tree and so on which have very small character and complex contours.Based on these, lots of works have been done by many scholars at home and abroad. After analyze and summarize the works of others, we have make research on several aspects which involves the local geometric properties of point cloud data, surface reconstruction and irregular objects rending technology. The specific major works involved are as follows:1. An improved algorithm for searching points' k-nearest neighbors is presented based on data division method. The valid scale size of grid is introduced by overall considering the range of data set, the total numbers of points, the numbers of nearest neighbors and the searching efficiency and then the data set space is divided into a set of uniform grids. Based on these, the average distance among data is calculated out take account of the point cloud density. Then we compared the size of the relationship between the valid scale size of grid and the new grid size calculated out based on the average distance. And then the final grid size will be chosen out to divide the data again. The division result makes the range to be searched much smaller and the searching speed much faster.2. Based on the RBF neural network's advantages of approaching no-linear function, powerful anti-noise and the capability of repair and so on to reconstruct surface from an unorganized cloud of points. We make use of the method of the correlation coefficient between points to self-adjusting the radius of the core function in the network. The algorithm's scheme and analyses are given and the proposed method was applied to the free surface reconstruction from scattered points in which involve noise. Compared with the traditional methods which used to decide the RBF model. The neural network can not only approach the surface with high precision but also have stronger anti-noise ability and the capability of repair and so on. And the smooth of surface which we got from experiments is good.3. The particle system combined with texture mapping technology is applied to rending the irregular object. We use the particle system's overall management and thinking process to simulate the flame, we make use of the random function to set the particle's initial position at the same time we analyzed and simplified the fire flame. The combination of the dynamic whirlpool field and the dynamic regional which can be controlled by people can improve the truth and the real-time simulation of the flame. Finally, we combined the quadrilateral surfaces with the texture mapping technology to further improve the flame drawing rate.At the last chapter, we conclude the thesis and discuss some existing defects and directions for future work.