Tree Segmentation of Airborne LiDAR Point Clouds Using a Graph-Based Algorithm

Forest monitoring using LiDAR has proven highly effective for obtaining detailed 3D point clouds, enabling individual tree analysis. This study introduces a novel segmentation algorithm designed for processing airborne LiDAR point clouds of forested areas. The proposed graph-based approach enhances the segmentation process by constructing voxelized clusters that form the basis for tree identification and isolation. The algorithm's performance was evaluated against three established methods: Dalponte's, Silva's, and Li's algorithms. Evaluation of the algorithm using metrics yielded the following results: recall = 0,778, precision = 0,809, and Fl-score = 0,742, indicating that while our approach does not consistently outperform all metrics, it demonstrates promising results with potential for further refinement. These findings suggest that our algorithm could contribute valuable insights into automated forest assessment and resource management as the methodology advances.