Automated productivity analysis of cable crane transportation using deep learning-based multi-object tracking

Abstract

The automated monitoring of construction equipment productivity has been a crucial research topic in intelligent construction, supporting refined construction management. This paper presents a vision-based monitoring method for automated productivity analysis of cable crane transportation in dam construction. It employs a deep learning-based Multi-Object Tracking (MOT) method to track the moving trajectories of crane buckets. Based on the trajectory data, the transportation productivity of cable cranes is calculated accurately. The MOT method integrates small object detection layers, tracklet information (short trajectory fragments), and global position relationships into the YOLO-DeepSORT framework to enhance tracking performance in the construction industry. Experimental results show improvements of 95.9% in IDF1 and 92.1% in MOTA on three long videos collected from dam construction sites. These results indicate that the proposed method captures moving trajectories accurately and analyzes transportation productivity effectively. • A vision-based monitoring method for automated productivity analysis is proposed. • Develop a deep learning-based multi-object tracking method for accurate tracking. • Address discontinuous situations using global position relationships and GNN models. • Address mutual occlusions of similar appearance objects using tracklet information. • Long-term accurate tracking and automated productivity analysis are achieved.