Abstract. Recent advances in photogrammetry and remote sensing have highlighted the advantages of three-dimensional (3D) point cloud data for accurately reconstructing different scenarios and environments, including forest and agricultural sites. Currently, LiDAR (Light Detection and Ranging) systems are widely used to acquire 3D data, offering high geometric precision and adaptability across various platforms. In comparison with conventional field surveys, information obtained by LiDAR systems provides greater spatial coverage and efficiency, particularly in large-scale applications. However, the automatic registration of point clouds in complex and irregular environments, such as forests, remains a challenge due to occlusions, repetitive patterns, and low overlap between scans. This paper proposes a coarse-to-fine point cloud registration approach designed for tree-dense environments. The method begins by processing each point cloud, acquired from different stations, to generate a Canopy Height Model (CHM). To determine tree center positions (xc, yc), the point clouds are sliced at breast height (1.30 ± 0.01 m), and the points corresponding to each tree cross-section are used to compute the center. These center features are subsequently matched using the relaxation labeling (RL) algorithm, which is based on probabilistic similarity and spatial relationships. From these corresponding trunk center pairs, it is possible to estimate an initial 2D transformation between the scans. This initial alignment is then refined using the Iterative Closest Point (ICP) algorithm to compute the final 3D transformation. Experiments using pairwise terrestrial scans with low overlap (< 30%) achieved a root mean square error (RMSE) of approximately 3 cm, without the use of artificial registration targets.
Silva et al. (Fri,) studied this question.