虚拟场景的视点质量评价与绘制优化问题研究

发布时间：2018-05-13 19:56

本文选题：虚拟场景 + 视点质量　；参考：《国防科学技术大学》2010年博士论文

【摘要】： 虚拟场景的视点质量评价与绘制优化是计算机图形学和虚拟现实技术领域中非常重要的研究课题,一方面对场景理解、科学可视化、三维模型检索等研究领域具有一定的理论指导意义;另一方面在虚拟战场、虚拟城市、影视动画和三维游戏等众多应用领域有着重要的实用价值。如何对虚拟场景的观察视点进行评价,并从视点观察质量角度出发对虚拟场景进行绘制优化,这一工作具有一定的困难和挑战,同时也是进一步提升虚拟场景观察和漫游绘制质量的必然要求。本文的研究工作主要涵盖了虚拟场景的视点质量评价、自动漫游路径生成、网格简化、地形与地物匹配等方面,对其中的一些理论和应用问题进行了深入分析和探讨。具体而言,论文的主要工作和贡献体现在以下几个方面: 提出了一种特征信息加权的视点质量评价方法。从信息获取的角度,将虚拟场景的特征信息和几何信息作为主要评价因素,对虚拟场景的视点质量进行评价。首先采用网格显著度对场景的特征信息进行度量,在此基础上对观察视点进行量化处理;然后根据建立的观察信道模型,定义了一个理想观察视点;最后将观察视点与理想视点之间的相似性测度作为观察视点的视点质量。实验对比表明本文算法对观察视点的评价结果较为客观,获得的最优观察视图具有较好的视觉效果。此外,根据图形硬件的高速并行特性,设计并实现了一种基于CUDA的视点质量计算方法,提高了视点质量的计算效率。提出了一种基于视点质量评价的自动漫游路径生成算法。自动漫游路径的生成包括最优视点集选取和路径生成两个关键步骤。对于最优视点集的选取,提出了两种不同的方法:一种是采用模糊C-均值聚类算法,根据聚类中心求解最优视点集;另一种是将观察视点集对应的虚视点的视点质量作为目标函数,并采用蚁群优化算法进行求解。在路径生成的过程中,设计了一种排序和平滑算法,解决了路径自交叉和拐点尖锐的问题。算法所生成的自动漫游路径光滑平整,且沿该路径观察时能够最大限度地获取场景信息,提高了观察和漫游绘制的效率与效果。提出了一种基于视点质量评价的网格简化算法。为了提高场景网格的简化质量,以观察视点的视点质量变化驱动场景网格的简化。算法首先在场景网格周围以最优分布或均匀分布方式布置观察视点;然后采用特征信息加权的视点质量评价方法计算三角形折叠引起的视点质量误差;接着将视点质量误差与二次误差结合定义三角形折叠的误差测度;最后根据误差测度确定三角形的折叠次序,从而完成场景网格的简化。算法能够保持场景网格中的显著区域,并能够对不可见三角形进行优先简化。提出了一种基于微分域网格变形算法的地形与地物匹配优化算法。为了解决虚拟场景中地形与地物的匹配失真问题,首先提出了一种改进的微分域网格变形算法,根据顶点的切平面中值坐标定义微分坐标;然后将细节编码和解码过程移植到GPU中,实现了一种图形硬件加速的微分域网格变形算法;最后根据地物的底面特征对地形网格进行形变处理。算法不仅实现了虚拟场景中地形与地物的自适应匹配,具有较好的匹配效果,而且由于将匹配问题转换为线性系统求解问题,因此具有较高的执行效率。以虚拟战场环境的相关应用为任务背景,设计并实现了一个优化漫游绘制原型系统——BOExplore,对虚拟场景的视点评价与绘制优化的研究成果和算法进行了应用和验证,并在相关项目中得到了应用。
[Abstract]:The quality evaluation and rendering optimization of virtual scene is a very important research topic in the field of computer graphics and virtual reality. On the one hand, it has a certain theoretical guiding significance for the research fields such as scene understanding, scientific visualization, 3D model retrieval and so on. On the other hand, virtual battlefield, virtual city, film and television animation and 3D. Many application fields, such as games, have important practical value. How to evaluate the view point of the virtual scene and optimize the virtual scene from the viewpoint of observation quality, this work has some difficulties and challenges, at the same time, it is also the necessary requirement to further improve the quality of virtual scene observation and roaming.
The research work of this paper mainly covers the quality evaluation of the virtual scene, the generation of automatic roaming path, the simplification of the grid, the matching of terrain and ground objects and so on. Some of the theories and application problems of the virtual scene are deeply analyzed and discussed.
A quality evaluation method of feature information weighting is proposed. From the angle of information acquisition, the feature information and geometric information of the virtual scene are used as the main evaluation factors to evaluate the visual point quality of the virtual scene. First, the feature information of the scene is measured by the grid saliency. On this basis, the observation view is carried out. According to the established observation channel model, an ideal observation view is defined. Finally, the similarity measure between the observation point and the ideal point of view is used as the visual point quality of the observation view. The experimental comparison shows that the evaluation result of the observation view is more objective, and the optimal view view has good view. In addition, according to the high speed and parallel characteristics of the graphics hardware, a method of calculating the quality of view based on CUDA is designed and realized, which improves the calculation efficiency of the view point quality.
An automatic roaming path generation algorithm based on view quality evaluation is proposed. The generation of automatic roaming path consists of two key steps: optimal view set selection and path generation. Two different methods are proposed for the selection of optimal view point set: one is to use fuzzy C- mean value clustering algorithm to solve the optimal view according to the cluster center. In the process of path generation, a sort and smoothing algorithm is designed to solve the problem that the path is cross and the turning point is sharp. The automatic roaming path generated by the calculation method is smooth and smooth. When the path is observed, it can maximize scene information and improve the efficiency and effectiveness of observation and roaming rendering.
A mesh simplification algorithm based on view quality evaluation is proposed. In order to improve the simplified quality of the scene grid, the view point quality changes are used to simplify the scene mesh simplification. First, the observation points are arranged in the optimal distribution or uniform distribution around the scene grid, and then the quality of view quality is weighted by the feature information. The evaluation method is used to calculate the quality error of the view point caused by the triangle folding, and then the error measure of the triangle folding is defined by combining the point quality error with the two error. Finally, the order of the triangle is determined according to the error measure, thus the simplification of the scene grid is completed. It can be seen that the triangle is simplified as a priority.
In order to solve the matching distortion of terrain and ground objects in virtual scene, an improved mesh deformation algorithm for differential domain is proposed, which defines the differential coordinates according to the median coordinates of the tangent plane in the vertex, and then encodes and decodes the details. In GPU, a graphic hardware accelerated differential domain mesh deformation algorithm was implemented. Finally, the terrain mesh was deformed according to the ground features of the ground objects. The algorithm not only realized the adaptive matching of terrain and terrain in the virtual scene, but also had better matching effect, and the matching problem was converted to a linear system for solving the matching problem. As a result, it has high efficiency.
With the application of virtual battlefield environment as the task background, an optimized roaming prototype system, BOExplore, is designed and implemented. The research results and algorithms of visual point evaluation and optimization of virtual scene are applied and verified, and they are used in the related projects.

【学位授予单位】：国防科学技术大学
【学位级别】：博士
【学位授予年份】：2010
【分类号】：TP391.41

【引证文献】