Abstract
Abstract
Occlusion and reflection phenomena in complex workpieces often cause phase information loss and point cloud " voids" in structured light 3D measurement. While dual-monocular systems can mitigate this issue through complementary viewing angles, they face the core challenge of point cloud misalignment between subsystems. To address point cloud misalignment in dual-monocular systems, this paper proposes a point cloud reconstruction method based on dual-monocular point cloud misalignment correction. Preprocessing achieves precise segmentation of foreground and background. Uniformly robust feature points are selected through a triple mechanism combining Monte Carlo sampling with foreground region validation, phase data integrity validation, and Euclidean distance constraints. Precise matching is achieved through polar line constraints and phase consistency, followed by transformation matrix solution. This enables high-fidelity acquisition of point cloud information in highly reflective and occluded scenes, along with accurate registration of point clouds from the dualmonocular system. Experimental results demonstrate that the corrected point cloud exhibits an 87% reduction in distance error and a 68% decrease in standard deviation. In occluded scenes, the number of valid points increases by 51% compared to other methods,while the standard deviation of point cloud density in highly reflective scenes decreases by 36%. These findings validate the accuracy, completeness, and stability of the proposed method.