【作者】 刘刊；

【导师】 李金宗；

【作者基本信息】 哈尔滨工业大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 月球探测车是用于复杂地形探测的移动机器人,是人工智能、自动控制、机构学、信息技术及计算机技术等多学科、高新技术的结晶,具有重大的研究价值和应用价值,受到了世界各国的重视。在月球车的各项关键技术中,立体视觉系统的主要功能是对月球表面复杂环境进行有效的感知和信息融合,完成图象的获`+取、像点匹配、障碍物识别与定位、三维场景重构和深度图生成等,以便实现月球车在未知环境里的自主导航和避障。在一定的计算能力下,视觉系统的速度和精度是一对矛盾。如何平衡两个性能是系统设计中的重要问题。针对这一问题,本文提出了月球车立体视觉的系统结构和设计方法。在保障月球车安全自主避障和导航的任务要求下,将视觉系统的基本性能与月球车的机动性能相结合,分析了视觉系统的各项性能,建立了一系列配置方程,利用这些方程选择合适的配置参数,达到优化性能参数的目的。双目视觉的精确标定是立体视觉算法的基础。针对月球车的应用背景,本文研究一种基于平面模板的摄像机在线标定方法,实现对月球车视觉系统的快速、精确标定。由于月球车立体视觉系统具有大畸变的特点,研究消除立体图象对的非线性畸变是必要的。通过双目视觉的外极线校正,可以将匹配的搜索空间由二维降低到一维。为此,本文研究一种紧凑的立体图象极线校正方法,根据摄像机标定的内外参数,对左右图象对进行极线校正。立体匹配是视觉算法的核心。本文首先对目前各种匹配算法进行分类,然后结合视觉避障导航的应用背景,突出应用环境对算法的准确性、鲁棒性和实时性的要求,对局部匹配算法和全局匹配算法进行比较研究。在局部匹配算法中对区域法的各种相似性测度的准确性以及匹配窗口的大小进行定性和定量的评估;在全局匹配算法中,研究动态规划方法和Graph Cuts方法,对这两种方法的准确性进行定量和定性的评估。为了提高结果的准确性和鲁棒性,对立体匹配进行预处理和后处理是必要的,包括消除左右图象亮度差异的LOG滤波、检测遮挡的左右一致性校验、及消除误匹配的blob滤波和中值滤波算法。为了研究比较各算法的性能,本文采用一种立体匹配算法的评估方法,对各算法结果进行定量的评估。更多还原

【Abstract】 Lunar Rover is an intelligent mobile robot to explore complex terrain, which is the product of artificial intelligence, automation, mechanism, information technology, computer science and other high-tech technology. It is of great value in research and application, and is paid great attention to all over the world. Of all the Lunar Rover key technologies, the main task of the stereo vision system is to achieve onboard perception and fusion of the complex lunar terrain, by capturing images, matching feature points, producing depth map, recognizing obstacles and reconstructing 3D scene, in order to navigate and avoid obstacle automatically in unknown environment.Response and accuracy conflict with limited computational resource. How to maintain the balance of these performances is important. This paper proposes a methodology for the system configuration and design on the rover stereo vision system. The paper analyzes performance of the vision system and presents principles for designing a stereo vision system for planetary rover by fusing main performance of a stereo-based system and the maneuverability of lunar Rover. Performance parameters are optimized, by applying the principles to choose suitable configuration parameters.Camera calibration is essential for stereo vision. A plane-based on-line calibration method is investigated , to calibrate cameras accurately and effectively. Deformation produced by short-focal and wide-angle lens should be undistorted. The searching space in stereo match is reduced from two dimension to one by rectification. A compact rectification method is studied. By applying the perspective matrix, the stereo image pair can be rectified.Matching is the core of stereo vision. Stereo matching can be divided into local (window-based) algorithms and global algorithms, which both have been investigated in this paper. For rover application, speed and robustness is considered, as the final system is intended to work in a dynamic environment under real-time condition. The topics of our research include correspondence calculating and window size in area-base algorithm, and dynamic programming and Graph Guts methods in global algorithms, also the evaluation of these更多还原