Computer vision-based process time data acquisition for offsite construction: Fid-Bau Portal

Computer vision-based process time data acquisition for offsite construction

Alsakka, Fatima / El-Chami, Ibrahim / Yu, Haitao / Al-Hussein, Mohamed

Abstract The application of computer vision in offsite construction for time data acquisition has proven valuable, but current methods pose a significant set-up burden. In the research presented in this paper, a computer vision-based process time data acquisition system (TiDA) for acquiring productivity-related data (productive time, cycle time, and process start-time) at the workstation level in panelized construction without the need for significant set-up efforts is proposed. The logic of the proposed system is demonstrated in reference to workstations on a wall production line in a lightweight wood panelized construction factory. Moreover, its promising performance is described in detail in a case application to framing operations in the factory. The mean absolute errors corresponding to the process start-time, productive time, and cycle time needed to frame 121 panels are found to be 0.72 min, 0.96 min, and 0.77 min per panel, respectively.

Highlights • Productivity data acquisition in offsite construction is challenging. • Computer vision is effective but requires excessive set-up effort. • The proposed TiDA measures productive-time, cycle-time, and start-time of processes. • TiDA uses YOLOv4 pre-trained on COCO dataset for object detection to acquire data. • TiDA acquires process time data with average errors of <1 min.