Registration method for weak-geometry point clouds from fringe projection profilometry by 2D matching and modulation-based adaptive weighting

Abstract

Point cloud registration remains challenging when the measured objects exhibit weak geometric features, where conventional geometric descriptors are often insufficient for establishing reliable correspondences. To address this issue, this paper proposes a coarse-to-fine registration framework that integrates two-dimensional image matching with threedimensional point cloud refinement. In the coarse registration stage, detector-free local feature matching with transformers is used to construct cross-modal correspondences, followed by a scale-invariant geometric consistency filtering strategy to suppress mismatches and improve the reliability of the estimated transformation. Note that the image for the coarse registration can be computed from the fringe patterns captured in the fringe projection profilometry. In the fine registration stage, a progressive modulation-weighted point-to-plane iterative closest point and normal iterative closest point scheme is adopted to improve local alignment accuracy. Experiments on a custom paper-sheet specimen and a textured scale vehicle model show that the proposed method achieves sub-millimeter registration accuracy, with a root mean square error of 0.107 millimeters and a mean absolute error of 0.011 millimeters, demonstrating its effectiveness for weak-geometry point cloud registration in fringe projection measurement.