【作者】 余洪山；

【导师】 王耀南；

【作者基本信息】 湖南大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 主动立体视觉技术是最接近于人类方式的视觉感知技术，也是计算机视觉获取三维信息最重要的方式。其原理是利用两个或多个摄像机在不同角度观察同一场景，通过获取立体图像对的视差，并进而实现恢复环境的三维信息。主动立体视觉技术广泛应用于自主式视觉导航、运动目标监视跟踪、工业加工的三维信息获取等领域。 本论文首先设计和开发了仿人多自由度立体双目视觉装置，该装置可水平旋转、俯仰、左右摄像机聚散度调整四个自由度伺服控制，并可实时在线调整摄像机参数。装置设计包括自由度控制电机及其驱动控制、摄像机同步控制、摄像机参数控制、基线长度选择等部分，并参考国内外同类设计和实际要求，实现了机械结构的优化设计。 在立体图像的特征提取中，本文提出了一种改进型Canny边缘检测算法实现图像精细边缘的检测，然后基于图像边缘梯度信息，采用分层并行迭代链码跟踪方法实现直线段的提取。在检测的直线段的基础上，根据角点和凸多边形的几何意义，实现了对图像中的角点和凸多边形区域等特征量的提取。 根据立体图像检测出的直线段、角点和多边形区域特征信息，并利用立体匹配过程的极线约束、连续性约束、唯一性约束、空间顺序约束等约束条件，实现了立体图像的特征匹配处理。 摄像机定标是实现三维信息恢复和极线约束的必要条件，因此本论文提出了一种立体双目摄像机的在线自定标方案。该方法基于立体摄像机的结构特点，并加以合理的假设，利用立体摄像机连续两次的运动序列所得的两对立体匹配图像，实现立体摄像机的定标。当聚散角和俯仰角在一定范围时，该方法具有很好的定标精度，但要求在运动过程中保持焦距参数保持不变。本论文还给出基于对应点和直线段信息的三维信息恢复的基本原理。 文章最后讨论了立体视觉技术在自主电动车系统的应用。更多还原

【Abstract】 Sharing mechanisms with biological vision systems, the active stereovision system is the primary way to obtain 3D information in machine vision field, whose applications span navigation of mobile robots, autonomous vehicles, surveillance, industrial process and so on.Firstly, This paper presents the design and development of active stereovision platforms and their integration into real-time closed loop active vision systems. The active stereovision system, built from off-the parts, allows accomplishing visual tasks independently like pan, tilt, vergence and camera parameters adjustment. Important parameters for the stereovision system such as the selection of motor resolution and size of baseline have been taken enough consideration. The strategies adopted for the design of the system, including mechanical design of platform, closed loop servo controller for motor of pan, tilt and vergence, camera parameters adjustment controller, and system integration, proved to be very successful.As for stereo matching, feature based matching method is preferred than intensity-based. Line segments, corners and polygonal regions are selected as features to be matched. Firstly, an improved Canny edge detection algorithm is represented to obtain thin and robust edges. Secondly, based on detected edge gradient of image, an iterative chain code tracing method with parallel hierarchy is introduced for line segment detection. Finally, combined with definition and geometrical structure, corners and convex polygonal regions are detected based on line segments. The use of these high level features, although helping to diminish matching ambiguities, does not completely resolve when the scene contains similar structures. The epipolar constrain, unique and continuity constrain, order constrain, spatial constrain and so on are employed for stronger support for correct feature matching pairs and, as a result, incorrect matching pairs can be largely eliminated.Camera calibration is an important prerequisite for retrieving 3D information of scene. Considered with peculiar configuration of stereo camera, some reasonable assumptions are introduced to obtain simplified stereo camera model. Based on the simplified form, a self-calibration method for moving stereo camera is represented in this paper, which calculates camera parameters through a successive motion sequence while camera focus remains invariable during the motion. The principle of retrieving3D information based on corresponding feature points and line segments is also discussed later.Finally, new generation autonomous vehicle based on stereovision navigation is highlighted as a future application.更多还原