ForestAlign: Automatic Forest Structure-based Alignment for Multi-view TLS and ALS Point Clouds

TL;DR

ForestAlign is proposed: an effective, target-less, and fully automatic co-registration method for aligning forest point clouds collected from multi-view, multi-scale LiDAR sources that can effectively integrate TLS-to-TLS and TLS-to-ALS forest scans.

Abstract

Access to highly detailed models of heterogeneous forests, spanning from the near surface to above the tree canopy at varying scales, is increasingly in demand. This enables advanced computational tools for analysis, planning, and ecosystem management. LiDAR sensors, available through terrestrial (TLS) and aerial (ALS) scanning platforms, have become established as the primary technologies for forest monitoring due to their capability to rapidly collect precise 3D structural information. Forestry now recognizes the benefits that a multi-scale approach can bring by leveraging the strengths of each platform. Here, we propose ForestAlign: an effective, target-less, and fully automatic co-registration method for aligning forest point clouds collected from multi-view, multi-scale LiDAR sources. ForestAlign employs an incremental alignment strategy, grouping and aggregating 3D points based on increasing levels of structural complexity. This strategy aligns 3D points from less complex (e.g., ground) to more complex structures (e.g., tree trunks, foliage) sequentially, refining alignment iteratively. Empirical evidence demonstrates the method's effectiveness in aligning scans, with RMSE errors of less than 0.75 degrees in rotation and 5.5 cm in translation in the TLS to TLS case and of 0.8 degrees and 8 cm in the TLS to ALS case, respectively. These results demonstrate that ForestAlign can effectively integrate TLS-to-TLS and TLS-to-ALS forest scans, making it a valuable tool in GPS-denied areas without relying on manually placed targets, while achieving high performance.

Figures (13)

• Figure 1: Alignment and accumulation of 5 multi-view TLS and ALS in New Mexico forest.
• Figure 2: Co-registration schematic. Two point-clouds, labeled source and target, are co-registered using our 3D structural complexity-based approach. On the left, point clouds are color-coded according to the parameters of a 3D plane approximation. Note that colors in the ground and tree trunks/branches appear more uniform compared to foliage, where colors are more randomly distributed or noisy. In the second column, we use a mixture of von Mises-Fisher (vMF) distributions (with the number of groups $K=3$) to group points according to their structural complexity, compactness, or informativeness. In the third column, we split the point-clouds according to the structural groups and assign a correspondence between scan groups. Finally, the right-most block co-registers in an incremental fashion by aggregating 3D points of increasing structural complexity, refining the estimated co-registration parameters.
• Figure 3: Pairwise alignment views of TLS-to-TLS co-registration. First row is an example result of pair-wise co-registration. (a) illustrates two scans miss-aligned, each in its own sensor coordinate system. (b-d) shows multiple views of the aligned point-clouds after applying our co-registration approach. Second row shows 5 aligned point clouds (using our approach), color coded each, distinctly.
• Figure 4: Robustness of the structural complexity-based algorithm against random alignment initializations in the TLS-to-TLS co-registration problem
• Figure 5: Pairwise TLS alignment in steep terrain.