Large-Range Reconstruction With Non-Pre-Calibrated Camera via MBM of Laser Referenced by Radial-Collinear-Features

Abstract

The 3-D reconstruction is of great importance for measuring, inspecting, monitoring, or understanding surface characteristics. Previous active target-based vision methods provide the reconstruction with more than three images for monocular vision, usually using extra external monocular, stereo, or multiview vision to expand the measurement range. This article aims to achieve the reconstruction from the initial image of monocular vision without precalibration. The subpixel centers on the target with groups of three collinear concentric circles in the radial pattern are obtained in the region of interest (ROI) by histogram of oriented gradient (HOG)–support vector machine (SVM), grouped by electromagnetic (EM), and extracted by the Zernike moment (ZM). The intrinsic parameters of the projection of the single image are modeled by the two projections of the general-direction infinity points generated from the collinear centers of the radial pattern. Moreover, the ratio of the cross ratios of CSI and CST features with infinity points is proposed to determine moderate images with small errors for reconstruction. The mutual benchmark model is presented to enlarge the effective measurement range without extra external observation instruments. Moreover, the reconstruction is modeled and further refined in SE(3) to achieve the orthogonality, accuracy, and large measurement range, which are proved by simulation and experimental verification.