A point cloud-vision hybrid approach for 3D location tracking of mobile construction assets

Abstract

Modeling as-is site condition and tracking the three-dimensional (3D) location of mobile assets (e.g., worker, equipment, material) are essential for various construction applications such as progress monitoring, quality control and safety management. Many efforts have been dedicated to vision-based technologies due to their merits in cost-effectiveness and light infrastructure compared to real-time location systems (RTLS). However, a major challenge of vision-based tracking is that it lacks 3D information and thus the results are sensitive to occlusion, illumination conditions and scale variation. To address this problem, this study presents a point cloud-vision hybrid approach to reconstruct and update the area of interest in 3D for scene updating and mobile asset tracking. Baseline 3D geometry information in point cloud is obtained at the start by Structure from Motion (SfM) using Unmanned Aerial Vehicle (UAV), given which mobile and static assets present in the scene are recognized and labeled. Based on 2D aerial isometric images capture by the UAV, labeled assets are automatically recognized and their locations are updated. The proposed approach was implemented in a field test and the results demonstrate that it was able to reconstruct the site and update the location of mobile assets accurately and reliably. Findings in this study indicate the proposed hybrid approach effectively augments the state-of-the-art in site modeling and asset tracking in construction.