Research On Viewpoint Quality Measuring And Rendering Optimization Of Virtual Scene

In computer graphics and virtual reality, viewpoint quality measuring and rendering optimization is an important research subject. On one hand, it can provide some theoretical guidance to many research areas, for instance, scene understanding, scientific visualization and 3D models retrieval. On the other hand, it has important practical value in some application areas, such as virtual battlefield, virtual city, video animation and 3D game. So the questions are brought forward: How to measure the viewpoint quality in virtual scene? How to optimize the scene rendering based on viewpoint quality? The work will come up against some difficulties and challenges to a certain extent, meanwhile it also can enhance the observation and exploration in virtual scene.The research of this thesis mainly focuses on viewpoint quality measuring in virtual scene, automatic exploration path planning, mesh simplification, terrain and ground features matching and so on. Some corresponding theory and application problems are analyzed and discussed thoroughly. In detail, the main contributions of this thesis are described as follows:A feature information weighted viewpoint quality measuring method is proposed. From the perspective of information acquiring, feature information and geometric information of virtual scene are taken as main measuring factors, and these measuring factors are used to calculate the viewpoint quality. Firstly, mesh saliency is used to describe the feature information of a scene, and the viewpoint is quantified on this basis. Then, according to the projection observation model, an ideal viewpoint is defined. Finally, the similarity between common viewpoint and ideal viewpoint is taken as viewpoint quality. Experimental results show that this algorithm can measure viewpoint quality objectively and can obtain an optimal view which has good visual effect. In addition, taking advantage of high speed and parallel characteristics of the graphics hardware, a CUDA based viewpoint quality computing method is designed which can improve the computational efficiency.An automatic exploration path planning method based on viewpoint quality measuring is proposed. The process of automatic path planning contains two key steps: optimal viewpoint set selection and path generation. Two different approaches to resolve the problem of optimal viewpoint set selection are proposed. The first one is based on fuzzy C-means clustering algorithm, and the optimal viewpoint set can be computed from the cluster centers. The second one takes the quality of virtual viewpoint, corresponding to a viewpoint set, as the objective function, and employs the ant colony optimization algorithm to solve this problem. In the process of path generation, a sorting and smoothing algorithm is designed to avoid self-intersection and abrupt turning point. The smooth path generated by this method can improve the efficiency and visual effect of observation and exploration, along which the information about the virtual scene can be acquired furthest..A mesh simplification method based on viewpoint quality measuring is proposed. In order to improve the quality of simplification, the changes of viewpoint quality are used to drive the process of simplification. Firstly, viewpoints are emplaced around the mesh uniformly or optimally. Secondly, the above viewpoint quality measuring method is used to calculate the change of viewpoint quality, which is brought by triangle collapse. Thirdly, quadric error and change of viewpoint are taken as the cost of triangle collapse. Finally, according to the collapse cost, the order of triangle collapse can be determined, and the mesh simplification can be accomplished. In the process of simplification, the salient area on the mesh can be preserved and the hidden triangles can be deleted in the first place.A terrain and ground features matching method based on differential domain mesh deformation is proposed. In order to solve the distortion of matching, a modified differential domain mesh deformation algorithm is put forward firstlly, which defines the differential coordinates of vertex by its mean value coordinates on tangent. Then, the process of detail encoding and decoding is transferred to GPU. Finally, on this basis, the terrain mesh is deformed according to the bottom characteristics of ground features. This algorithm achieves good matching, and converts a matching problem to a linear system solving problem, which makes its implementation more effective.A prototype system named BOExplore is designed, which takes the virtual battlefield as background. Methods about viewpoint quality measuring and rendering optimization in virtual scene proposed in this thesis are applied and validated in this system. Some parts of this system have been applied to correlative projects.