【作者】 丁志安；

【导师】 张丽艳；

【作者基本信息】 南京航空航天大学 ， 航空宇航制造工程， 2007， 硕士

【摘要】 工业生产中经常需要对产品进行三维表面外形和几何尺寸的测量。目前的三维测量方法都需要专门的测量设备且存在着种种限制。基于视觉的测量技术作为一种非接触式的测量手段,具有速度快、精度高、易于自动控制等优点。本文从计算机视觉的几何理论出发,研究采用一个数码相机,以自由拍摄方式获得场景的多帧数字图像,从中获取被测物体三维信息的方法。本文的主要研究内容包括在给定相机内参数和相机内参数未知的两种情况下,同时反求各次拍摄的相机位姿和三维点坐标;多帧数字图像中特征曲线的提取和匹配;相机位姿和内参数已知情况下的三维曲线结构重建。对于自由拍摄得到的多幅图像,本文首先构建测量矩阵并通过矩阵分解算法将测量矩阵分解为射影空间下的相机运动和物体形状,接着求解一个满足欧氏重建条件的非奇异矩阵,上升为欧氏空间下的相机运动和物体形状。在相机位姿确定后,运用Snake提取图像特征曲线,利用各幅图像间对应点的极线约束,建立了不同图像间同名曲线对上点的对应关系,研究并实现了基于两幅图像的三维曲线结构重建算法。在理论研究的基础上,开发了基于单数码相机的三维曲线重建系统。完成了显示器、摩托车覆盖件等样件的测量建模,取得了好的测量效果。更多还原

【Abstract】 Current methods of three-dimensional (3D) measurement have various kinds of disadvantages. The technology for 3D measurement based on computer vision is a non-contact measuring way, which provides a rapid, high-precision, maneuverable measurement method. This thesis deals with 3D reconstruction from views which are obtained by only one hand-held camera on the basis of computer vision.The main contents of this thesis include: 1) reconstruction of the camera pose (with known or un-known intrinsic parameters) related to each image, together with the 3D coordinates of the marked points; 2) the marked curves extraction and correspondences among multiple digital images; 3) 3D curves reconstruction with camera poses estimated.The measurement matrix is constructed and the projective reconstruction is obtained by decomposing the matrix. Then a non-singular matrix is evaluated, which satisfies the conditions of Euclidean reconstruction. This upgrades projective reconstruction to Euclidean reconstruction.The active contour model is used to extract the image curves. The correspondence of the images curves is estimated based on the theory of epipolar geometry after poses of the camera are estimated. Then a 3D curve reconstruction algorithm is put forward.Software for 3D curve measurement is developed on the basis of above research. Some models were successfully reconstructed. The result indicates that the proposed method is practical and efficient.更多还原