Computer vision and deep learning-based automation for rebar counting and reporting on-site

Abstract

Quality control on construction sites remains complex and predominantly relies on manual effort, particularly for tasks such as rebar counting, which are time-consuming and prone to errors. Although a large number of images are captured daily during the construction process, these visual data are mainly archived and not yet effectively utilized for automated quality control. Recent advances in deep learning, especially in computer vision, offer promising opportunities to automate the rebar counting process directly onsite. This paper proposes a comprehensive framework based on deep learning techniques to automate rebar counting, aiming to enhance both the efficiency and accuracy of construction quality control. The proposed framework consists of four main stages: (1) Data Collection, involving the capture and organization of rebar images; (2) Rebar Detection, applying advanced computer vision models to accurately identify rebars; (3) Automated Rebar Counting; and (4) Real-Time Report Generation. To ensure optimal detection performance, several popular object detection models, including Faster-RCNN, YOLOv8, YOLOv11, YOLOv12, and RF-DETR, were trained and compared. Evaluation results show that YOLOv8 outperformed the others, achieving a mean Average Precision (mAP@50) of up to 98% while operating in real-time. The proposed solution significantly reduces manual labor, minimizes counting errors, and improves the overall effectiveness of quality control procedures in construction projects.