Neighborhood constraint extraction for rapid modeling of point cloud scenes in large-scale power grid substations

Abstract

3D laser scanning is widely used in modeling industrial scenes, involving crucial steps like point cloud (PC) data preprocessing and extracting modeling targets from scattered points. Extracting target PC data from larger scenes is more challenging than from smaller settings. To efficiently obtain high-quality scene data and modeling targets, we introduce an innovative approach with the adaptive density-based voxel grid filtering algorithm and a probability statistics histogram method during preprocessing. We also propose a novel method for automatic extraction of target PC data using the adjacent feature plane constraint (AFPC) clustering technique. Initially, we capture height characteristics of the target object's PC through experiential and statistical height attributes. Points corresponding to each height are then clustered based on the PC's density distribution using the density-based spatial clustering of applications with noise algorithm. The intersection of these results optimizes the position of the target object's extraction bounding box, achieving seamless automatic object extraction. Experimental results validate the effectiveness of our preprocessing methodology, with F1 scores of 97.7 % and 97.3 % for the 220 kV and 500 kV areas, respectively. Furthermore, our novel extraction method demonstrates the ability to autonomously and directly extract electrical equipment from substation PC data.

Keywords

• Point clouds data
• Voxel grid filtering algorithm
• Density-based spatial clustering
• Object extraction
• 3D laser scanning