【作者】 王俊；

【导师】 吴谨；

【作者基本信息】 武汉科技大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 计算机立体视觉的主要研究内容是由多幅二维的平面图像恢复出被摄物体的三维坐标,它具有非接触性测量、实施较为简易等特点。作为计算机视觉的一个重要分支,立体视觉是一个多学科交叉的领域,吸引着大批包括视觉生理、心理、物理、数学以及计算机科学等多学科的研究人员运用不同的技术手段对之进行深入的研究,不仅具有重要的理论意义,并且在现代工业、国防、医学、空间技术等领域有着广阔的应用前景。一个完整的立体视觉系统可以划分为图像获取、摄像机标定、特征提取、立体匹配、3-D信息恢复以及后处理六个模块。本文首先对计算机视觉的基本概念和视觉过程的特性进行介绍,接着从摄像机标定、特征提取、以及立体匹配3个方面展开研究。摄像机标定即确定摄像机参数,是计算机视觉研究中由图像进行三维重建不可或缺的重要步骤。本文在分析理想的摄像机投影模型的基础上,利用旋转矩阵的正交性和摄像机参数间的约束关系,设计摄像机标定系统,计算摄像机的内部参数与外部参数,实验结果表明,该算法减少了运算量和运算时间,标定结果具有较高的精确度。对灰度图像进行特征提取等预处理过程是进行图像配准的前提,本文对图像的噪声消除、图像增强进行了简单介绍,对不同类下的特征特性和提取方法进行了全面的研究和比较,通过实例对算法进行试验,得出各种算法的处理结果。在图像特征提取与描述的基础上,即可进行图像的配准工作。本文详细论述了特征基元选择、匹配策略、约束条件以及相似性测度等立体匹配的主要组成部分,提出了一种基于图像特征控制点匹配方法和一种基于区域相关性的立体匹配方法,通过实例对算法进行分析,得到较好的匹配效果。更多还原

【Abstract】 Stereo vision is to reconstruct the three-dimensional coordinates of objects from several 2D images. It has the characteristics of non-contact measure and simple actualization. As an important branch of computer vision, stereo vision is a discipline-crossing field and attracts many researchers who adopt different technologies, including vision psychology, physic, mathematics and computer science and so on, to research. The technology has important academic significance and a wide application prospect in the areas national defense, medicine, spatial technology.The integrated stereo vision system generally includes six parts: image acquisition, camera calibration, feature extraction, stereo matching, 3D information recovery and post-processing. This paper firstly introduces the basic concept of computer vision and characteristics of the visual process. Then we mainly discuss camera calibration, feature extraction, and stereo matching.Camera calibration is to ascertain parameters of camera and is the important step of 3D reconstruction from images in computer vision research. Based on analyzing ideal camera projection model, this paper utilizes the orthogonality of rotation matrix and the constraint relationship among camera parameters to design the camera calibration system and compute the inner and outer parameters of the camera. The experiment results show that the algorithm could reduce computation burden and has high precision.The pre-processing is the precondition for image matching. This paper briefly introduces image denoise, image enhancement, researches and compares features’characters and extraction methods in detail, and do experiments to validate the algorithms.Image matching is based on feature extraction and image description. This paper discusses the important constituent part of stereo matching, including feature units selection, matching strategy, restriction, and similarity measurement. Aiming at stereo images with different specialties, this paper presents a matching algorithm based on image feature control points and a matching algorithm regional relativity. The matching results have good performance.更多还原