In engineering construction projects, rebar spacing measurement requires significant manual labor with low efficiency. This paper proposes a new intelligent rebar spacing measurement method based on the YOLOv8-GB model to save the workforce and improve efficiency. This method collects images of rebars to be measured using a binocular camera, utilizes the proposed YOLOv8-GB model to extract rebars from the scene, and achieves spacing measurement. The system is deployed on the NVIDIA Jetson TX2 NX for on-site portable measurement and can run in real-time at 24 frames per second. Experimental results show that the improved YOLOv8-GB network, compared with the YOLOv8n network, increased Recall, Precision, mAP@0.5, and mAP50-95 by 0.6 %, 5.5 %, 2.3 %, and 7.6 %, respectively. The measurement system built with YOLOv8-GB achieved an average absolute error of ± 1.7 mm, ±2.1 mm, and ± 2.7 mm for rebar spacing measurements on three different ground textures, with average relative errors of 0.85 %, 0.93 %, and 1.32 %, meeting engineering requirements. Compared to the measurement system built with YOLOv8n, the average absolute error decreased by 37.0 %, 8.0 %, and 25.0 % under the three different ground textures, while the average relative error decreased by 36.1 %, 8.8 %, and 23.7 %, respectively.
Read full abstract