Answers are grounded in this paper and cite their sources.
Aerial and ground-based laser scanning platforms provide complementary three-dimensional, high-precision data for characterizing forest structure, yet accurate registration of cross-platform point clouds remains challenging in complex forest environments. Existing methods rely either on object-level features (e.g. tree stems), which are often incomplete across platforms, or on local geometric descriptors that are unreliable given self-similarity, occlusion, and density asymmetry inherent in forest point clouds. To address these limitations, we introduce MultiVoxelReg, a multi-scale voxel-based framework for automatic cross-platform forest point cloud registration. Instead of depending on semantic objects or geometric keypoints, MultiVoxelReg leverages voxel-level statistical properties that are more robust to noise and partial data loss. Forest structure is represented using a 12-dimensional voxel descriptor encoding distributional, positional, and geometric characteristics of points within voxels. Multi-scale structural information is captured through hierarchical voxelization, enabling coarse-to-fine correspondence matching. This is further coupled with a dynamic-threshold Iterative Closest Point refinement stage and a global initial alignment module that accommodates large pose discrepancies. We evaluate the proposed method on 15 forest plots, including 10 UAV-terrestrial LiDAR plots from a public dataset and 5 UAV-backpack LiDAR plots collected in this study, spanning diverse forest structures, terrain conditions, and initial misalignments. The results demonstrate consistent robustness and accuracy, achieving mean translation and rotation errors of 5.23 cm and 0.13° for UAV-terrestrial data, and 17.00 cm and 0.29° for UAV-backpack data. These results highlight voxel-based statistical representation as an effective and generalizable approach for cross-platform point cloud registration in structurally complex environments.
No takes yet. Share an insight, caveat, or question.
Song et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: