High-accuracy point cloud registration for 3D shape measurement based on double constrained intersurface mutual projections

The accuracy of 3D shape measurement directly determines the quality and reliability of intelligent manufacturing, and point cloud registration is the key factor. However, the dislocation of discrete point clouds in the overlapping area seriously reduces the registration accuracy. This paper proposes a new high-accuracy registration method based on double constrained intersurface mutual projections. First, the initial registration set is built by mutual projections between similar local regions, then the final registration set is determined by the rigid transformation consistency constraint, finally, high-accuracy pairwise registration is realized. On this basis, the K-means clustering is further fused to achieve multiview global optimization. On Stanford dataset, both of the pairwise and multiview registration error significantly decreased. In the experiment with GoScanG1 scanner, the surface error of pairwise registration was reduced by 2.2%; that of the multiview registration was reduced by 42.5%, which shows the advancement of the new method in actual measurements.

Automated summary

This paper proposes a high-accuracy registration method based on double constrained intersurface mutual projections for 3D shape measurement. It effectively reduces the dislocation of discrete point clouds in the overlapping area, improving registration accuracy and reliability.