基于图像的单视点场景建模

【作者】 石睿；

【导师】 李华；

【作者基本信息】 中国科学院研究生院（计算技术研究所） ， 计算机应用技术， 2001， 硕士

【摘要】 基于图像的建模和绘制是近年来兴起的一种倍受关注的场景建模和绘制方法,也是目前虚拟场景漫游领域中的主流技术。其最大的特点就是克服了传统的基于几何建模和绘制的不足,可以实现真实感很强的场景漫游效果。在此背景下,本文对于基于图像的单视点场景建模问题进行了探讨和研究。首先,本文综述了IBM、IBR以及单视点场景建模的各自技术特点和最新研究进展,对所涉及的射影几何、仿射几何及线性摄像机模型等相关理论和概念进行了介绍。在此基础上,本文研究了特定场景的单视点建模以及纹理平滑等问题。本文的主要内容包括:1.针对某些特定的场景,尤其是含有人造建筑物的场景,结合具体实例提出了一种基于图像的单视点场景建模方法。该方法以射影几何、仿射几何为理论基础,以线性摄像机为模型,在满足两个前提假设条件下,利用图像中灭点、灭线等几何信息并通过灭线变换、对应点变换等数学方法,求得图像点与实际空间点之间的变换关系,进而恢复出场景的几何信息并为场景建模。与一般的IBM技术相比,该方法最大的特点就是计算量小、建模过程简单。2.　由于受到图像分辨率的影响,从场景图像中获取的纹理图像往往分辨率不够,因此漫游过程中纹理易出现马赛克现象,从而影响了场景的真实感。针对这一点,本文提出了一种基于Catmull-Clark细分曲面的纹理平滑算法。该算法具有计算简单、易实现、图像放大时相邻像素的RGB值能够平滑过渡等特点。平滑后的纹理图像在场景漫游中仍保持了良好的连续性和光滑性,取得了比较理想的效果。　　虽然从目前来看,基于图像的单视点场景建模无论在理论框架的建构,还是在具体问题的解决能力上还处于比较初步的阶段,还有大量的工作需要做,但是从长远来看,基于图像的单视点场景建模仍然是人们所关注的领域,而且只要多借鉴其它领域的技术(如纹理映射技术、绘制技术等)来完善自身的不足,它仍将是一个充满巨大发展潜力的领域。同时,这也是本文对今后工作的一个展望。更多还原

【Abstract】 Image Based Modeling & Rendering is a fascinating technique of modeling and rendering in recent years, and now it is also the essential technique in the field of virtual navigating. Furthermore, it’s most remarkable that IBMR can overcome the deficiency of traditional Geometry Based Modeling and Rendering technique, so IBMR can realize photorealistic modeling and navigating. Under this background, the thesis researches and discusses the problem of scene modeling based on single view image, which people have focused on for many years. Firstly, we summarize the speciality and new development of IBM、IBR and single view modeling respectively. At the same time, we introduce some key concepts and theories which are touched in this thesis. In this framework, the thesis presents a modeling method of special scene based on single view and an algorithm of texture smoothing, and so on. The main content of this thesis is as follows:1. With a given in-door scene, this thesis puts forward a modeling method of special scene based on single view image, which applies to some scenes, especially scenes containing man-made buildings. This method is based on Projective Geometry and Affine Geometry with the linear camera model. Satisfying two premise conditions, this method can obtain the transforming matrix between image pixels and points in the world by some geometry information, such as vanishing point、vanishing line, and some math transformations, such as Vanishing-Line-Transformation、Correspondences-Transformation. So the 3D scene can be modeled according to these geometry information of the scene. This method has the property of smaller computations and simpler modeling procedure.2. This thesis presents an algorithm of texture smoothing based on Catmull-Clark subdivision surfaces. Influenced by image resolution, the scene texture image always has lower resolution so that mosaic appears easily in course of virtual navigating. Because Catmull-Clark subdivision surfaces have better property of continuity and smoothness, the algorithm of texture smoothing based on Catmull-Clark subdivision surfaces has such characters as simpler computations, better continuity and smoothness in color space, and so on. Finally this algorithm get a more perfect result.更多还原