An automated vision-based approach for structural dynamic identification

This paper presents an automated computer vision-based procedure for assessing structural displacements, aiming to perform modal identification and evaluate structural responses under dynamic loading. The proposed algorithm is designed to extract structural displacements from recorded videos by temporally tracking predefined points on the structure. The research addresses key methodological challenges in monitoring large structures, including detecting the relatively small displacements characteristic of structural vibrations under environmental conditions, correcting for typically unavoidable perspective distortions and camera movements, reconstructing the 3D displacements from single-camera videos and identifying global mode shapes. The effectiveness of the proposed procedure is first demonstrated through experiments on a laboratory-scale frame under controlled conditions. Its applicability to real-world scenarios and ability to identify global mode shapes are further evaluated using experimental data from a full-scale footbridge. In both cases, comparisons between vision-based monitoring results and reference measurements demonstrate the strong performance of the method. The proposed approach achieves an accuracy below one-tenth of a pixel, corresponding, in the investigated experimental case study, to a metric accuracy of approximately 0.1 mm at a 60 m camera-to-target distance and 0.03 mm at 25 m. This confirms the method’s capability to provide reliable measurements for real-world structural monitoring applications.