Abstract
The application of LiDAR technology in extracting individual trees and stand parameters plays a crucial role in forest surveys. Accurate identification of individual tree trunks is a critical foundation for subsequent parameter extraction. For LiDAR-acquired forest point cloud data, existing two-dimensional (2D) plane-based algorithms for tree trunk detection often suffer from spatial information loss, resulting in reduced accuracy, particularly for tilted trees. While cylinder fitting algorithms provide a three-dimensional (3D) solution for trunk detection, their performance in complex forest environments remains limited due to sensitivity to parameters like distance thresholds. To address these challenges, this study proposes an improved individual tree trunk detection algorithm, Random Sample Consensus Cylinder Fitting (RANSAC-CyF), specifically optimized for detecting cylindrical tree trunks. Validated in three forest plots with varying complexities in Tianhe District, Guangzhou, the algorithm demonstrated significant advantages in the inlier rate, detection success rate, and robustness for tilted trees. The study showed the following results: (1) The average difference between the inlier rates of tree trunks and non-tree points for the three sample plots using RANSAC-CyF were 0.59, 0.63, and 0.52, respectively, which were significantly higher than those using the Least Squares Circle Fitting (LSCF) algorithm and the Random Sample Consensus Circle Fitting (RANSAC-CF) algorithm (p < 0.05). (2) RANSAC-CyF required only 2 and 8 clusters to achieve a 100% detection success rate in Plot 1 and Plot 2, while the other algorithms needed 26 and 40 clusters. (3) The effective distance threshold range of RANSAC-CyF was more than twice that of the comparison algorithms, maintaining stable inlier rates above 0.9 across all tilt angles. (4) The RANSAC-CyF algorithm still achieved good detection performance in the challenging Plot 3. Together, the other two algorithms failed to detect. The findings highlight the RANSAC-CyF algorithm's superior accuracy, robustness, and adaptability in complex forest environments, significantly improving the efficiency and precision of individual tree trunk detection for forestry surveys and ecological research.