【作者】 徐玮；

【导师】 吴玲达； 张茂军；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2007， 博士

【摘要】 动态虚拟环境的建模与绘制是当前基于图像的建模与绘制(简称IBMR)技术领域中的难点和研究热点。本文在折反射全景技术的基础上,提出了一种动态虚拟环境建模与绘制方法,构造的动态虚拟环境不仅能够实现空间域中一定范围的自由漫游,而且当动态虚拟环境中仅存在周期性、类周期性或统计性运动场景时,可以在视觉上表现动态虚拟环境无限时间的连续变化。根据IBMR技术构造虚拟环境的通用流程,本文的主要工作包括样本数据获取(第二章)、动态虚拟环境建模(第三章)以及动态虚拟环境绘制(第四、五、六章)三个部分,主要内容概述如下:一、本文以折反射全景作为样本数据,样本采集方面主要研究折反射全景获取与柱面展开算法。在深入研究折反射全景成像技术的基础上,设计并实现了一种折反射全景成像装置;推导了将该成像装置拍摄的环状全景图像展开为柱面全景图像的光路跟踪展开算法;针对光路跟踪展开算法中存在的计算量大、实时性差等不足,提出了同心圆环近似快速展开和八项对称重用快速展开两种快速展开算法。二、研究基于折反射全景的动态虚拟环境建模方法。以折反射全景数据作为样本,提出了一种基于“动静”分离原则的动态虚拟环境表示模型,将动态虚拟环境分割为动态区域和静态区域分别描述,然后通过时空关联进行整合。针对该模型,提出了“动静”分离的建模框架,并分别提出了基于折反射全景图像移动拍摄系统的静态区域虚拟环境自动建模方法、基于端点同步全景视频的动态区域虚拟环境建模方法以及对静态区域和动态区域进行空间关联的方法。三、研究基于折反射全景的动态虚拟环境绘制方法及其所涉及的关键技术。本文是基于柱面视图合成来实现动态虚拟环境绘制的,而对极几何估计是进行视图合成的基础,因此,动态虚拟环境绘制所涉及的关键技术主要包括对极几何估计和柱面视图合成,本文按照先关键技术,后绘制方法的顺序进行本部分内容的研究。1)对极几何估计。提出了一种基于特征匹配的对极几何自主估计方法,首先自动获取图像对之间的匹配点集,然后以此为样本,估计可描述对极几何约束关系的基础矩阵,为了提高匹配点集的精度,提出了一种基于双向最大相关和视差约束的特征匹配方法,为了提高基础矩阵估算的鲁棒性,提出了一种基于群体智能的基础矩阵估计算法。2)柱面视图合成。提出了先自主估计对极几何约束关系,再图像校正和视图插值的分步式视图自动合成算法框架,提出了基于平面图像极线校正和基于柱面图像极线校正两种柱面视图合成方法,并对两种算法的差异及其实验结果进行对比和分析,选出了更适合本文动态虚拟环境绘制要求的柱面视图合成方法。3)动态虚拟环境绘制。首先提出了基于柱面视图合成的动态虚拟环境生成方法来生成当前观察视点和观察时刻的动态虚拟环境,然后提出了基于透视投影变换实现对当前观察视点和观察时刻动态虚拟环境快速显示的方法。综上所述,本文围绕动态虚拟环境的构造,依次研究了折反射全景获取与柱面展开、动态虚拟环境建模和动态虚拟环境绘制。更多还原

【Abstract】 Modeling and rendering of dynamic virtual environment is the technically difficulty and the focus of research in Image-based modeling and rendering (IBMR). In this paper, a dynamic virtual environment modeling and rendering method based on the technology of catadioptric panorama is presented. The dynamic virtual environment constructed by the method mentioned above can not only realize free movement in space to some extent, but also show the visual continuous change of the virtual environment with infinite time when there is only repetitive, quasi-repetitive or statistic moving scenes.According to common flow that constructs the virtual environment with the IBMR technology, the main working in this paper includes sample capturing (chapter 2), dynamic virtual environment modeling (chapter 3) and dynamic virtual environment rendering (chapter4, 5, 6). The following is the abstract of main content:1. In this paper, catadioptric panorama is selected as sample data, and the research on sample capturing mainly focus on catadioptric panorama capturing and cylindrical unwrapping algorithm. Based on the deep research of catadioptric panorama imaging technology, a catadioptric panorama imaging device is designed and realized. An optical tracking unwrapping algorithm that unwraps the circular panorama captured by this imaging device into cylinder panorama image is presented. In order to overcome the shortcoming such as large amount of computation and poor property on real time in the above method, this paper presents a fast approximation concentric circles unwrapping algorithm and a fast eight direction symmetric reuse unwrapping algorithm.2. The dynamic virtual environment modeling method based on catadioptric panorama. Using the data of catadioptric panorama as sample, A dynamic virtual environmental expression model based on dynamic and static separating strategy is presented. It divides the dynamic virtual environment into dynamic and static area to be described separately, and then incorporates these two area with the spatial relationship. For this model, the paper presents a dynamic-static separating modeling framework, an automatic static area virtual environment modeling method based on moving capture system of the catadioptric panoramic image, a dynamic area virtual environment modeling method based on synchronic omni-directional video and a spactial relating method to the static and dynamic area.3. The dynamic virtual environment rendering method based on catadioptric panorama and its relative key technology. In this paper, the realization of dynamic virtual environment rendering is based on cylindrical view synthesis. The epipolar geometry estimation is the basis of view synthesis, so the relative key technologies of the dynamic virtual environment rendering mainly include epipolar geometry estimation and cylindrical view synthesis. In this paper, the sequence of the research is the key technology first and the rendering method later.1) The epipolar geometry estimation. An epipolar geometry estimation method based on feature matching is presented. it is the main idea of the above method that matching points set of image pairs is automatically gained to estimate the fundamental matrix, which can describes the constraint relationship of the epipolar geometry. In order to improve the precision of the sets of matching points, a feature matching method based on bidirectional maximum correlation and parallactic constraint is presented. To improve the robustness of the fundamental matrix estimation, an algorithm for fundamental matrix estimation based on swarm intelligence is presented.2) Cylindrical view synthesis. An automatic multi-step view synthesis framework is proposed, the main idea is at first automatically estimating the constraint relationship of the epipolar geometry, and then image rectification, at last view interpolation. Two cylindrical view synthesis methods are presented, one is based on planer image epipolar rectification and the other is based on cylindrical image epipolar rectification. According to comparison and analysis of experimental results from those two methods, the more suitable cylindrical view synthesis method is chosen.3) Dynamic virtual environment rendering. At first, a dynamic virtual environment generating method based on cylindrical view synthesis is presented to render the dynamic virtual environment of the current observe viewpoint and observe time. And then, the paper presents a method based on perspective projection, which quickly display the dynamic virtual environment of current observe viewpoint and observe time.Above all, this paper focuses on the core question about how to construct dynamic virtual environment,and in sequence explore catadioptric panorama capturing and cylindrical unwrapping, dynamic virtual environment modeling and rendering.更多还原