Computer Vision-based Algorithm for Deformation Monitoring in Bridge Construction Research

The aim of this paper is to present a methodology for the detection of deformations in bridges during construction. Using the YOLOv5 model, a detection and localization model was trained to identify checkerboard targets. Relevant features were then extracted from the detected target patterns. To determine changes in target poses, the correspondence was established between 3D world coordinates and 2D pixel coordinates. Utilizing this correspondence, the real-time target poses were calculated, providing valuable insights into the magnitude and direction of deformations. The proposed visual deformation measurement method is dependent on the target and the camera, enabling real-time and cost-effective measurements of target displacement. The experimental results demonstrated the efficacy of our approach, which exhibited remarkable accuracy and robustness.