Multisource LiDAR Point Cloud Registration Method Based on Progressive Optimization of Triangular Mesh Similarity in Forest Environments

The registration of LiDAR point clouds across multiple platforms in forest environments is a crucial step in forest inventory. However, registering point clouds from different sources in forests poses challenges, such as unstable features for alignment, limited robustness across varying forest conditions, and uncertainties in integrating point clouds obtained from diverse LiDAR platforms. In order to address these challenges, this study introduces a method for registering multi-source LiDAR point clouds based on progressive optimization of triangular mesh similarity. This paper first leverages several spatial geometric features for stable registration, generated based on relative distances among neighboring trees and angles between the growth direction and the vertical zenith. These developed geometric features remain consistent and stable across different scales of LiDAR source point clouds, enhancing the applicability of LiDAR point cloud data registration across diverse platforms. Subsequently, a progressive determination and optimization strategy for registration was devised to reduce sensitivity to tree positioning accuracy in different sources of LiDAR point clouds and improve the robustness of the proposed method, particularly in complex forest environments. Furthermore, a similarity matrix was constructed to facilitate point cloud data registration across different platforms. With this matrix's assistance, point clouds from terrestrial laser scanning (TLS) and unmanned aerial vehicle laser scanning (ULS), TLS and mobile laser scanning (MLS), as well as backpack laser scanning (BLS) and ULS can be correctly registered. In summary, the key contributions of this work are threefold: firstly, this paper leveraged several spatial geometric features for stable registration; Secondly, a progressive optimization strategy was introduced for improving robustness across varying forest conditions; Lastly, a similarity matrix-based triangular mesh matching framework was developed that enables cross-platform registration (TLS/ULS/MLS/BLS). The effectiveness of the proposed method was evaluated using 14 plots in three distinct forest sites. Experimental results demonstrate that the method yields satisfactory registration performance across various combinations of point cloud sources, including TLS with ULS, TLS with MLS, and BLS with ULS. The average residual distance between reference and estimated coordinates using the calculated transformation matrix was found to be 0.1589 m for the 14 plots. The average rotation and translation errors were calculated as 0.0056 degrees and 0.3470 m, respectively, confirming the successful registration outcomes achieved by the proposed method. Comparative analysis against five classical registration methods revealed that the proposed method consistently outperformed its counterparts across main accuracy indicators considered.