Evaluation of Octree-Based Segmentation (OBS) Method to Seperate Ground Point Based on the Handheld Laser Scanner Data

Accurate digital elevation models (DEMs) are crucial for effective forest planning and management. Various methods exist for generating DEM data, with handheld mobile laser scanners being the most efficient and precise approach. The raw point clouds obtained from these scanners require several preprocessing steps, one of which involves separating ground and non-ground points. Errors in this part of the process can lead to the generation of an inaccurate digital model with high uncertainty and errors. Various algorithms, such as voxel-based segmentation, simulation filters, and deep learning-based approaches, have been developed for this purpose. This study evaluates the performance of the OBS algorithm in automatically separating ground points from non-ground points in handheld laser scanner data. The study area comprised five different sections within the Karaj Botanical Garden, covering a total area of 7.2 hectares. These areas contained forest stands characterized by heterogeneous structures and multi-story tree layers. Data were acquired using a handheld GeoSLAM laser scanner. To generate a reliable reference for evaluating the algorithm's results, ground points were manually separated. The performance of the algorithm was evaluated by comparing it with the manually separated ground truth using statistical metrics, including Matthew's correlation coefficient, Kappa coefficient, and Intersection over Union (IoU). The statistical metrics across the five study areas demonstrated the effectiveness of the OBS algorithm in separating ground points from non-ground points, with Matthew's correlation coefficient, Kappa coefficient, and IoU values of 0.895, 0.891, and 0.902, respectively. Additionally, the optimal voxel size for the algorithm was determined to be within the range of 15 to 22 centimeters.We conclude that the OBS algorithm, when configured with optimal input parameters, provides high performance in automatically separating ground points from non-ground points, especially in heterogeneous forested environments. The importance of configuring the optimal input parameters is also highlighted.