Three-dimensional reconstruction method for sparsely textured surfaces via freely moving line-structured-light-based binocular stereo vision

Three-dimensional (3D) reconstruction using line-structured-light-based stereo vision is one key technique in the field of computer vision measurement. However, most existing stereo vision methods are static and have limited measurement range. To overcome the defect in static stereo vision, a 3D reconstruction method for sparsely textured surfaces via freely moving line-structured-light-based binocular stereo vision is proposed in this paper. The light stripes projected on the surface are captured by left and right cameras, and the pixel coordinates of the light strips are extracted by the Steger algorithm. The 3D coordinates of the light stripes at each position of the cameras are calculated based on the principle of binocular intersection measurement. Subsequently, at least six feature points are required to estimate the coarse relative rotation and translation between adjacent positions, and the corresponding precise values are computed based on minimization of pixel reprojection errors. The camera pose at each position is ultimately obtained through step-by-step backward propagation by using the relative rotation and translation between all adjacent positions. Thus, the 3D shape of the surface is obtained quickly once the light strips captured at different positions are unified to the reference position. Moreover, two metallic surfaces with different geometric configurations are used for the 3D reconstruction test, and the 3D shapes of the metallic surfaces are reconstructed effectively, which proves the generalizability of the proposed method to different surface types. To evaluate the accuracy of 3D reconstruction, the 3D point cloud of one metallic plane is fitted into a plane, and the average vertical distance of the 3D point cloud from the fitting plane is regarded as the evaluation criterion. The experimental results reflect that the average vertical distance in each experimental trial and the average value of the standard deviation of vertical distance do not exceed 1.52 and 0.97 mm, respectively.