High-accuracy multi-view three-dimensional measurement of large-scale components using complementary gray code and LM-ICP registration

Abstract To address the issues of limited field of view, complex surface reflection and easy accumulation of multi-view stitching errors in the manufacturing and inspection of large-scale components, a high-accuracy multi-view three-dimensional (3D) shape measurement method for large-scale components is proposed in this paper. A positive–negative complementary gray code phase-shifting structured light coding method is developed in the single-view 3D reconstruction step of the proposed method. Combined with the robust pixel classification and the adaptive Gaussian filtering strategy, it effectively enhances the robustness of absolute phase unwrapping under complex reflection conditions and reduces the single-view geometric reconstruction error. An Levenberg–Marquardt-optimized iterative closest point (LM-ICP) point cloud fine registration algorithm integrating common region constraint and normal vector consistency constraint is constructed, which significantly suppresses the accumulation of registration errors caused by mismatches in multiple stitching processes, in the multi-view point cloud stitching step. Meanwhile, the influences of the coding mode, the number of multi-views, the camera resolution and the exposure time on the measurement accuracy of the proposed method are analyzed. Measurement experiments on dual-ball-bar standard artifacts and typical industrial large-scale components verify that the proposed coding method improves the single-view reconstruction accuracy by about 24.5%, and the LM-ICP point cloud fine registration algorithm increases the overall scale measurement accuracy by about 29.6%. The system achieves stable sub-millimeter measurement accuracy under multi-view conditions.