基于未标定序列图的形貌测量方法

提出一种形貌测量方法,从未标定序列图中同时提取出特征点的三维坐标和摄像机的运动矩阵,实现对目标物体的三维测量。首先用齐次坐标表达了空间点的投影变换矩阵,并根据该矩阵推导出投影比例式,运用投影深度和投影矩阵之间的循环相关性进行迭代运算,得到比普通双目视觉更加精确的深度数据。其次对分解得到的摄像机运动矩阵和空间点坐标施加度量约束。运用基于因式分解的归一化算法,恢复出欧氏空间里的摄像机运动矩阵和空间点坐标。基于这种方法,实现了曲面标记点的坐标测量,得到了三维形貌数据,误差小于0.16mm。

A Shape Measurement Method Based on Uncalibrated Image Sequence

In this paper a novel reconstruction algorithm for extracting the three-dimensional coordinates of feature points in a structure and the movement matrix of video camera at the same time is proposed to realize three-dimensional shape measurement based on uncalibrated sequence of an arbitrary number of images. The scaled projection matrix was firstly discussed and expressed by homogenous coordinates of spatial points. The definition of projection ratio was derived from the matrix. Taking advantage of the projections of 3D points onto an image sequence, iterative projective depths were calculated, the depth data with higher accuracy than the binocular vision were obtained. The projective solution was then converted to a Euclidean one by enforcing metric constraints. A factorization-based normalization algorithm was presented to generate the Euclidean reconstruction and the intrinsic parameters. Results of reconstruction of regular marks on a non-planar surface by this method are presented in this paper. Three dimensional shape data were analyzed with absolute errors under 0.16mm.