A Canopy–Gap Boundary-Based Method for Multiplatform Forest Point Cloud Registration

Le résumé fourni par la source

The integration of LiDAR point clouds acquired from airborne, terrestrial, and mobile platforms can improve the characterization of forest spatial structure. However, robust cross-platform registration remains challenging because of occlusion, viewpoint differences, uneven point density, and the limited repeatability of individual-tree features in complex forest environments. To address these challenges, this study proposes an automated canopy–gap boundary-based registration method for multiplatform forest point clouds. First, canopy points are extracted using an adaptive height-difference threshold. Canopy and gap boundary polygons are then generated through buffer-union polygonization. Subsequently, multifeature polygon descriptors, geometric gating, global one-to-one assignment, and spatial-consistency filtering are used to establish reliable boundary correspondences for coarse registration. Finally, trimmed iterative closest point (ICP) is applied to refine the alignment. Experiments were conducted in six forest plots with different stand structures and point-cloud acquisition platforms. The proposed method achieved coarse-registration RMSE, MAE, median error, and P95 error values of 0.0847–0.1757, 0.0625–0.1258, 0.0472–0.1221, and 0.1671–0.3635 m, respectively. After ICP refinement, these errors decreased to 0.0500–0.1250, 0.0366–0.0958, 0.0267–0.1004, and 0.1073–0.2686 m, respectively. Compared with manual registration and two published baseline methods, the proposed method reduced the average RMSE by approximately 0.07 m in coarse registration and 0.03 m after fine registration. These results indicate that canopy–gap boundary objects can provide stable and interpretable constraints for ALS–TLS/MLS registration, offering an effective solution for multiplatform forest point cloud fusion and subsequent forest structural parameter extraction.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.