Highly accurate three-dimensional measurement of large structures using multiple stereo vision with improved two-step calibration algorithm

Abstract

A multiple stereo-based three-dimensional (3D) measurement method of large-scale structures is presented, which replaces time consuming and laborious 3D measurement procedure of laser-based conventional equipment. To have 3D reconstruction accuracy with millimeter-precision, two step calibration algorithms are improved. Specifically, we use 2D epipolar constraint, using fundamental matrix determined by matched points, to new combination of objective function for intrinsic parameter calibration. We additionally propose an onsite structure parameters calibration algorithm using different epipolar constraint of measurement points in images of large ship blocks as well as reprojection error of checkerboard, to bypass the limitation of the coverage of checkerboard in large-scale measurement. The actual 3D measurement of the three ship blocks is conducted, and the 3D reconstruction accuracy is validated by comparing the 3D coordinates measured by total station. The experimental results showed that the mean 3D reconstruction error ranged from 6 to 14 mm.