Cross-grid laser reconstruction method based on stripe splitting and coarse-to-fine matching

In conventional parallel multi-line laser three-dimensional (3D) reconstruction systems, the matching of the laser stripe center point (LSCP) relies on the laser-plane equation. Consequently, the matching accuracy is susceptible to deformations of the laser plane and degrades as the number of laser stripes increases. Moreover, the unidirectional distribution of the laser stripe (LaS) in the image for parallel multi-line setups increases the number of registration steps and prolongs scanning. To address these issues, a trinocular cross-grid laser (CLS) reconstruction method is proposed. First, the trifocal tensor (TFT) and the reference angle for CLS splitting are estimated via system calibration. Next, using the Hessian-matrix normal vector and the minimum directed boundary distance, the CLS is split into unidirectional LaS, and LSCP coarse matching is performed subject to epipolar geometry, disparity, and spatial consistency constraints. Fine matching is then performed via TFT-based point transfer to determine the correct matching triplet from the set of coarse-matched candidates, followed by reconstruction using the left and right cameras. Finally, global registration of local-view point clouds is realized using circular retroreflective markers affixed to the object. Experiments show that, in a 15-line CLS system, the matching rate (MRa) and matching accuracy rate (MAR) reach 97.4561% and 99.1072%, respectively; the reconstruction errors for a two-dimensional (2D) plane and a 3D sphere are less than 0.037 and 0.095 mm, respectively; and the system runs at approximately 12 fps with GPU acceleration. The proposed method alleviates the dependence of multi-line laser reconstruction systems on the laser-plane equation and provides, to our knowledge, a new method for line-laser reconstruction.