<italic>E</italic>-<italic>L</italic>₀: Advanced Surface Segmentation of LiDAR Point Clouds in Open-Pit Mine Stepped Terrain

Abstract

Ensuring the stability of open-pit mine slopes is crucial for safe and efficient mining operations. To analyze slope stability and assess geological disaster risks, LiDAR point-cloud data are widely used to create high-precision 3-D models. However, the existing segmentation methods, which are mostly designed for urban or indoor environments, struggle with the complex terrain of open-pit mines that includes both natural variations and artificial structures, such as benches and slopes. To address this challenge, we propose a new point-cloud segmentation method based on enhanced L0 gradient minimization (E-L0), specifically tailored for open-pit mines with benched topography. First, a normalized spatial metric is used to create a supervoxel set that preserves boundary features, thereby reducing the computation and handling density differences. Next, an adjacency graph is built, and the E-L0 generates initial planes. Finally, global energy optimization is applied to refine and merge these planes into a complete surface set. Given the lack of public benchmark datasets for open-pit mines, our method was tested on manually labeled data. It achieves average F1-scores of 74.7% for structural segmentation and 80.2% for boundary delineation when processing both airborne and vehicle-mounted LiDAR data. This method supports slope stability monitoring, 3-D reconstruction, and estimating quantities for earthwork in open-pit mining.

Keywords

• Benched topography
• energy optimization
• L0 gradient minimization
• point cloud
• supervoxel