Real-time assessment of rebar intervals using a computer vision-based DVNet model for improved structural integrity

Abstract

Structural durability is critical for building and civil engineering safety, wherein the arrangement and distribution of reinforcing bar (rebar) is crucial. Improperly aligned rebar impacts bearing capacity, whereas uniform spacing optimally distributes loads, reducing stress. We introduce a computer-vision based Deep Vision Net (DVNet) model for real-time evaluation of rebar placement. A customized dataset is prepared in an environmental setup and augmented to address overfitting issues. This research conducts a comparative analysis of the learning performance exhibited by the proposed DVNet model against several other pre-trained models, such as Mask-RCNN and YOLOv5. The proposed DVNet model is built on a customized DeepCNN architecture, achieving a commendable precision of 88.6 % and recall of 89.3 %. Utilizing the DVNet model, the real-time assessments of rebar placements were performed at various spacing intervals. Experimental results demonstrate that the DVNet-based model excels at ensuring the structural arrangements of the rebar intervals.

Keywords

• Rebar
• Structural health
• Convolution neural network
• Segmentation
• Deep neural network