In complex outdoor scenes, the precise registration of airborne LiDAR and ground-based LiDAR data can significantly enhance 3D observation capabilities, which is crucial for the construction and development of digital twins for cities. Geometric heterogeneity and coverage differences between cross-source point clouds often lead to ambiguous feature correspondences and poor registration performance, hindering the complete expression of multidimensional spatiotemporal data. We propose an incremental registration strategy with vertical and horizontal alignment. Vertical alignment quickly corrects positional deviations in the vertical direction. Horizontal alignment is based on point cloud projection raster images, where we propose a geometry-consistency-enhanced dynamic sampling RANSAC method that integrates multidimensional 3D geometric features to construct consistency weights, thereby overcoming the limitations of a single constraint. Furthermore, we develop a multi-scale geometric constraint adaptive-weight optical flow method that jointly captures initial keypoint correspondences and computes the initial transformation pose through an adaptive weight energy function and cross-scale consistency verification. Finally, we provide a fine registration algorithm with normal similarity weighting constraints to achieve precise fusion of point clouds. Multiple experimental results demonstrate that the proposed method achieves high accuracy, robustness, and generalizability with low computational cost and insensitivity to initial pose conditions. Compared to mainstream registration methods, such as GROR and EJRGF, this method demonstrates superior performance, significantly enhancing its spatiotemporal data observation capabilities in areas such as real-world 3D reconstruction.
Xu et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: