PSI - Issue 84

Giorgia Ghirelli et al. / Procedia Structural Integrity 84 (2026) 1047–1054

1049

2. Fundamental Concepts of Vision-Based Monitoring Vision-based structural monitoring builds on the fundamental assumption that, for a stationary camera, any change in the apparent position of a target between consecutive video frames corresponds to the displacement of the associated structural point. In the proposed methodology, this concept is combined with the use of high-contrast checkerboard targets, which enable robust feature detection and sub-pixel accuracy of features tracking. The developed workflow is designed to ensure consistent image quality, correct lens distortions, and provide reliable displacement measurements in real-world units. The designed pre-processing procedure employs the MATLAB Image Processing and Computer Vision Toolboxes (R2024), which allow to import the video, convert color images to greyscale, extract frames and remove lens distortion using camera parameters obtained through a dedicated calibration procedure. This step ensures that geometric inconsistencies introduced by the optical system do not affect the accuracy of the subsequent analysis. To reduce computation effort, a region of interest (ROI) is defined to isolate the relevant portion of each target. Within the defined ROI, each video frame is represented as a matrix of pixels characterized by 2-D coordinates and RGB intensity values. For checkerboard-pattern targets, the corners of the squares provide distinctive image features due to the sharp intensity gradients between adjacent black and white regions. A feature-point matching procedure is implemented: by computing the local image gradient and analyzing the RGB intensity variation across the pixel grid, the coordinates of the corners are detected with sub-pixel accuracy. Once the corners are identified in each frame, their average position is computed for displacement tracking. A pixel-to-metric conversion is therefore required to obtain displacements in physical units. The scaling factor is determined by comparing the known geometric dimensions of the target with the corresponding pixel distances between adjacent corners, evaluated separately in the horizontal and vertical directions. The dynamic displacement time history is finally obtained by computing the frame-by-frame change in target position with respect to its reference configuration. By repeating this process for each target installed on the structure, the displacement time histories of multiple sections can be reconstructed from the same video sequence. The methodology allows for the extraction of both global and local dynamic characteristics, providing a versatile and efficient tool for structural monitoring campaigns based on consumer-grade imaging equipment. 3. Case study: RC railway arch bridge over Panaro river The proposed vision-based dynamic monitoring approach was applied to a reinforced concrete railway bridge located in Savignano sul Panaro, Northern Italy. The structure crosses the Panaro river and it is composed of five identical spans, each measuring 39.45 m in length. Every span follows the same structural scheme, consisting of two reinforced concrete arches with eliminated horizontal thrust through tie elements and a suspended deck supporting the railway track. At their upper nodes, the arches are linked by eight transverse beams that provide lateral stiffness and contribute to the overall stability of the superstructure. Beneath the arches, the deck consists of a continuous RC slab supported by transverse beams spaced 2.63 m apart. These beams are suspended from the arches by reinforced concrete tie rods.

Fig.1. Longitudinal view of the bridge.