Efficiency Evaluation of Sampling Density for Indoor Building LiDAR Point-Cloud Segmentation.

Area of Science:

• Computer Vision
• Robotics
• Geomatics Engineering

Background:

• Indoor LiDAR point-cloud semantic segmentation accuracy is heavily influenced by sampling density, creating an accuracy-efficiency trade-off.
• Current density selection methods are often heuristic and lack standardized protocols, limiting quantitative guidance for practical applications.

Purpose of the Study:

• To develop a unified evaluation framework for analyzing the impact of sampling density on LiDAR point-cloud semantic segmentation.
• To provide reproducible, model-agnostic guidance for optimizing density in indoor mapping and Scan-to-BIM (Scan-to-Building Information Modeling) workflows.

Main Methods:

• A standardized protocol was used to evaluate three representative backbones (PointNet, PointNet++, DGCNN) augmented with a Point Transformer module.
• The framework employed isotropic voxel-guided uniform down-sampling and a decision rule integrating accuracy sufficiency, efficiency, and accuracy-density curve analysis.
• Experiments were conducted on indoor point clouds, contrasting scan-derived data with BIM-derived counterparts.

Main Results:

• The study quantified the accuracy-runtime trade-off, identifying an engineering-feasible operating band of 1600-2900 points/m², with a robust setting around 2400 points/m².
• Planar components showed saturation at moderate densities, while beam components were more sensitive to down-sampling.
• The proposed framework offers reproducible guidance by isolating density effects and standardizing evaluation protocols.

Conclusions:

• The research provides quantitative, reproducible, and model-agnostic guidance for scan planning and compute budgeting in indoor mapping.
• The findings are crucial for optimizing resource allocation and improving the efficiency of Scan-to-BIM processes.
• Establishing a standardized approach to density selection enhances the reliability and applicability of LiDAR-based indoor mapping solutions.