应用灭点标定的立体视觉自由曲面三维重建

Three dimensional reconstruction of stereo vision free-form surface based on vanishing-point calibration

为了实现对自由曲面的三维空间数据的测量与重建,建立了立体视觉自动测量系统,并对该系统的非线性标定及立体图的匹配方法进行了研究。基于灭点理论法进行了系统的非线性标定,采用投射光斑法增加自由曲面的图像特征;提出了一种基于方向和距离约束的分片排列编码方法,对投影光斑进行排列编码。分别在左右图像上根据彩色点的位置关系自动确定一个初始匹配点和4个匹配方向,然后对其余点按照分片排列法进行自动排列编码。对石膏像的面部进行了三维测量和重构实验,实验结果表明:像物尺度比为0.08 mm/pixel时,空间点三维坐标测量的均方差为0.05mm,可以达到亚像素级测量精度;基于方向和距离约束的分片排列编码法,可以准确地进行投影点的编码,并能够快速、有效地实现彩色图像特征点的全自动匹配。

In order to realize the automatic 3D measurement and reconstruction for a free-form surface, an automatic measuring system for the stereo vision is established and its adopted algorithms such as the nonlinear calibration for the measuring system and the match algorithm for stereo images are investigated. A nonlinear calibration method based on vanishing-point theorem is applied to calibrate the system and the image features of a free-form surface are enhanced by projecting arrays of dots. Then, an array coded algorithm based on direction and distance restrictions is presented to code projecting dots. After setting an original matching point and four original matching directions automatically based on the positions of red points, all other white dots viewed both in two images can be coded by the proposed method and the images in two directions can be matched automatically. The experiments of 3D measurement and reconstruction for a plaster model are undertaken, and the experimental results show that when scale is 0.08 mm/pixel, the square error of 3D measurement for dots is 0.05 mm, which can reach a sub-pixel measurement precision. Moreover, the projective points on a picture can be coded precisely and the matching between two images can be performed effectively by the array code method based on direction and distance restrictions.