三维场景的多视点感知算法研究

发布时间：2018-01-10 10:03

本文关键词：三维场景的多视点感知算法研究　出处：《中国科学技术大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着计算机技术的飞速发展,三维模型数据库的资源共享得以实现。为方便用户快速浏览并查阅三维模型,数据库往往被要求提供一个或一组预览图去表征一个三维模型。一组高辨别度的预览图可帮助用户快速辨别其对应的三维模型,相对地,一组低辨别度的预览图则会导致用户的判断出现差错。特别地,在家装设计完成后,室内设计师往往需要向用户提供一组整体效果图去感知整体家装风格。因此如何根据场景信息自适应地得到用户指定视点个数的最优多视点集合,使得其不仅能够表征三维场景,而且能够给予用户最大程度上的感知是一个重要且有意义的问题。在三维模型感知领域,研究人员更多地关注于单一三维模型的视点选择问题,而其中少量关注于复杂三维场景模型的相关工作也仅处理最优单视点的整体感知或局部感知问题。针对现有工作并未考虑三维场景多视点整体感知的问题,本文首次研究了三维场景多视点感知算法,以帮助人们尽可能地感知整个三维场景。针对现有工作中的视点度量算法并不适用于三维场景整体感知的问题,本文定义了一个新颖的视点感知度量函数,保证了优化视点能够覆盖足够多的几何特征的同时,更加贴近人们的视觉感知习惯。针对输入的三维场景模型,本文首先利用三维场景模型中几何与结构信息定义了基于模型本身的度量函数,其次利用视点对应观察视图的构图美学信息定义了基于观察视图的度量函数,进而加权组合两类度量函数并得到了一个涵盖几何信息、结构信息、美学信息等感知元素的视点感知度量函数,保证了其对应的最优视点在能够覆盖足够多显著几何信息的基础上更加贴近人们的视觉感知习惯。基于该视点感知度量函数与用户指定视点个数,本文构建了一个多视点整体优化模型,保证了多视点协同感知。为避免陷入局部最优,本文利用经典的模拟退火算法对优化模型进行多视点整体优化求解,得到最优多视点集合。为验证本文算法的有效性与合理性,本文对实验结果进行定量与定性的分析。通过与相关工作中度量函数的定量比较,证明了本文视点感知度量函数的有效性。通过与优秀室内装修设计师人为指定的最优视图集合进行定性比较,证明了本文算法的合理性。实验结果表明本文算法对不同类型的三维场景模型都可以在保证足够多的显著几何特征的前提下自适应地得到符合人们视觉感知习惯的最优多视点集合。
[Abstract]:With the rapid development of computer technology, the resource sharing of 3D model database can be realized. Databases are often required to provide one or a set of previews to represent a three-dimensional model. A set of high-resolution previews can help users quickly identify their corresponding 3D models, relatively. A set of low-resolution preview plans can cause errors in the user's judgment. In particular, after the home design is complete. Interior designers often need to provide users with a set of overall effects to perceive the overall home decoration style. Therefore, how to adaptively get the optimal set of multiple views according to the scene information. It is an important and meaningful problem that it can not only represent 3D scene, but also give users maximum perception. Researchers pay more attention to the viewpoint selection of a single 3D model. A small amount of work focused on the complex 3D scene model only deals with the global perception or local perception of the optimal single view. The existing work does not consider the problem of multi-view global perception of 3D scene. In order to help people perceive the whole 3D scene as much as possible, the multi-view perception algorithm of 3D scene is studied for the first time in this paper. For the existing work, the viewpoint measurement algorithm is not suitable for the problem of overall perception of 3D scene. In this paper, a novel viewpoint perception metric function is defined to ensure that the optimal viewpoint can cover enough geometric features and be more close to people's visual perception habits. This paper firstly defines the measurement function based on the model itself by using the geometry and structure information in the 3D scene model, and then defines the measurement function based on the observation view by using the composition aesthetic information of the viewpoint corresponding observation view. Then we combine the two kinds of measurement functions and obtain a viewpoint perception measure function which covers the sensing elements such as geometry information, structure information, aesthetic information and so on. It ensures that the corresponding optimal viewpoint is closer to people's visual perception habits on the basis of covering enough significant geometric information. Based on the view perception metric function and the number of points of view specified by the user. In this paper, a multi-viewpoint global optimization model is constructed to ensure multi-viewpoint cooperative perception. In order to avoid falling into local optimum, the classical simulated annealing algorithm is used to solve the multi-viewpoint global optimization of the optimization model. To verify the validity and rationality of the proposed algorithm, the experimental results are analyzed quantitatively and qualitatively. It proves the validity of the viewpoint perception measure function in this paper. It is qualitatively compared with the optimal view set of the excellent interior decoration designer. The experimental results show that the proposed algorithm can adapt to the visual perception habits of different types of 3D scene models, while ensuring sufficient significant geometric features. The set of optimal multi-view points for.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】