A Container-Aware Volumetric Framework: Poisson Reconstruction for Accurate Stockpile Measurement Using Lidar Point Clouds in Confined Spaces

Accurate stockpile volume estimation in confined industrial environments remains challenging due to complex geometries and environmental clutter. Traditional 2.5D elevation models often fail to capture critical three-dimensional features, while manual methods prove labor-intensive and unsafe. This paper presents a container-aware volumetric framework integrating Poisson Surface Reconstruction (PSR) with LiDAR point clouds to enable automated, precise volume measurements in GNSS-denied spaces. Our method leverages predefined 3D container models to filter environmental noise and performs spatial Boolean operations to isolate target stockpiles from structural interferences. Key innovations include a two-stage geometric filtering process combining axis-aligned bounding box (AABB) tree-based point cloud clipping with normal-constrained adsorption, followed by density-aware Poisson reconstruction. A mesh segmentation pipeline with morphological classification further ensures robust stockpile identification. Validated through a large-scale case study at a coal-fired power plant covering 44,500 m², the framework demonstrated sub-1% relative precision in two warehouse measurements, with reduced stability in occlusionprone areas. The proposed 3D mesh-based approach outperforms conventional 2.5D models in representing steep slopes and overhangs, while maintaining compatibility with GIS platforms for large-scale geospatial monitoring.