A clustering-based automatic registration of UAV and terrestrial LiDAR forest point clouds

Unmanned aerial vehicle laser scanning (ULS) and terrestrial laser scanning (TLS) provide complementary, nondestructive approaches to acquire three-dimensional forest structure information. Registration of their point clouds enables the reconstruction of complete vertical structure of forests. Current registration methods are primarily designed to register different TLS scans and thus are not applicable to ULS-TLS registration directly. In this study, the proposed method first generated multi-layer tree maps from ULS and TLS data using hierarchical clustering, then extracted Fast Point Feature Histograms (FPFH) features for each cluster point based on spatial relationships in the tree maps. After that, a point-to-point matching strategy was used to obtain the transformation matrix of each layer between ULS and TLS trunk point clouds and the best matrix from all layers was finally selected for fine registration. The algorithm was validated in 40 sample plots in Guangxi province of China. Our findings indicated that both high-density ULS and TLS data generate accurate tree maps compared to manually counted tree number, with a Concordance Correlation Coefficient (CCC) of 0.961 and 0.973, respectively. The proposed method performed well in registration accuracy and time efficiency, and achieved a higher matching score (0.945 > 0.928) and lower RMSE (0.144 < 0.151) than manual registration. The average registration time per sample plot of 600 m2 was 48.9 s, with 19.4 s dedicated to coarse registration. This research highlights the potential of clustering-based registration methods for effectively aligning ULS-TLS point cloud data in forests, laying the foundation for further technological advancements in forest vertical structure reconstruction.