During the construction phase, it is crucial to measure the pier settlement displacement and deformation of large-span prestressed concrete bridges to ensure the construction safety. Traditional contact measurement techniques, however, are time-consuming and labor-intensive, while noncontact vision sensing method has poor precision and robustness in complex construction environments, such as the object occlusion and varying light conditions. To address these issues, this study proposes a noncontact vision-based deformation measurement method for large-span rigid concrete bridges. Specifically, (1) a deep learning network was first employed to eliminate the adverse effect of complex backgrounds and varying ambient light in captured images on target detection results and an adaptive displacement extraction algorithm without a human-computer interaction process was developed to automatically extract target coordinates and pier settlement displacement by a dual camera system; (2) an enhanced vision-based deformation measurement methodology for large-span prestressed concrete rigid-frame bridges under construction scenarios involving personnel movement and mechanical operations that induce partial or complete target occlusion. The robustness and efficacy of the proposed method have been thoroughly verified through field tests on a prestressed concrete rigid-frame bridge during the symmetrical cantilever casting process. The results demonstrate that our proposed method greatly minimizes deformation anomalies due to object occlusion and efficiently captures deformation from targets of various shapes, such as circular and chessboard patterns. This method demonstrates significant potential for accurately measuring multipoint deformations of large-scale bridges in complex construction environments, thereby providing essential data for bridge safety assessment and construction strategy decision-making.

Bridge engineering; displacement measurement; computer vision; deep learning; intelligent construction