LiDAR-Based Segmentation of Tall Vegetation

This paper presents a comparative analysis of various segmentation methods for extracting tall vegetation objects from LiDAR data, including clustering-based methods, voxelization strategies, and Canopy Height Model (CHM)-based segmentation. First, an overview of Light Detection and Ranging (LiDAR) technology is provided, focusing on its underlying principles and the unique advantages it offers in forestry management, such as detailed canopy structure mapping. Subsequently, an extensive experimental evaluation is conducted on real forest datasets, where each segmentation technique is assessed in terms of accuracy using Intersection over Union (IoU) metric. Through these experiments, the paper highlights key trade-offs between accuracy and runtime performance, revealing that methods leveraging voxelization can achieve high IoU values while remaining computationally feasible. The findings underscore the importance of selecting an appropriate segmentation algorithm for forestry applications, as it can significantly impact the fidelity of canopy measurements and downstream forest management decisions. The experimental results demonstrate that combining voxelization with DBSCAN algorithm yields superior segmentation performance compared to other evaluated CHM-based methods.