Principal Axes Descriptor for Automated Construction-Equipment Classification from Point Clouds

Abstract

Recognizing construction assets (e.g., materials, equipment, labor) from point cloud data of construction environments provides essential information for engineering and management applications including progress monitoring, safety management, supply-chain management, and quality control. This study introduces a novel principal axes descriptor (PAD) for construction-equipment classification from point cloud data. Scattered as-is point clouds are first processed with downsampling, segmentation, and clustering steps to obtain individual instances of construction equipment. A geometric descriptor consisting of dimensional variation, occupancy distribution, shape profile, and plane counting features is then calculated to encode three-dimensional (3D) characteristics of each equipment category. Using the derived features, machine learning methods such as k-nearest neighbors and support vector machine are employed to determine class membership among major construction-equipment categories such as backhoe loader, bulldozer, dump truck, excavator, and front loader. Construction-equipment classification with the proposed PAD was validated using computer-aided design (CAD)–generated point clouds as training data and laser-scanned point clouds from an equipment yard as testing data. The recognition performance was further evaluated using point clouds from a construction site as well as a pose variation data set. PAD was shown to achieve a higher recall rate and lower computation time compared to competing 3D descriptors. The results indicate that the proposed descriptor is a viable solution for construction-equipment classification from point cloud data.