PSI - Issue 37

Francisco Barros et al. / Procedia Structural Integrity 37 (2022) 880–887 Barros et al./ Structural Integrity Procedia 00 (2019) 000 – 000

883

4

Images of the reference state, i.e. the bridge’s surface with no load on it, were then acquired from both cameras on the 3D rig simultaneously, as well as from the ground floor camera. After this, a static load was applied by a group of seven people standing on the bridge, near the section being monitored, as shown in Fig. 1, but such that they would not appear on the camera images. Images were then acquired of the bridge ’s surface under this load ed condition. The reference images are shown in Fig. 3; the deformed images are nearly identical to the references since the displacements are so small.

(a)

(b)

Fig. 3. Reference images acquired by: (a) camera A; (b) camera B.

All images were acquired manually, using the manufacturer’s software for the 2D case and, for the 3D camera rig, an application developed to visualise and acquire images from both cameras simultaneously. Even though the trigger for acquisition was sent at the same time, the images obtained were often a few seconds out of sync, therefore only a static load, held for a long amount of time, was feasible to evaluate. 3. Image processing 3.1. Camera calibration The camera calibration for the 3D setup was carried out using the Zhang calibration algorithm (Zhang, 2000). Second-order and fourth-order radial distortion were computed, but higher-order radial distortion, tangential distortion and pixel skew were assumed to be zero. On the 2D setup, it was impossible to capture the intended surface such that the camera sensor was parallel to it, due to physical obstacles, so a correction of the distortion caused by perspective had to be implemented. For this purpose, one image (Fig. 4) of a small calibration pattern pressed against the side of the bridge was acquired. An optimal projective transformation was computed such that these corners defined squares of equal size, i.e., the real geometry of the pattern. Applying this transformation to the DIC images makes the distances and in-plane displacements on the corrected images proportional to the real distances and displacements. The pixel to millimetre correspondence after correction is obtained through the knowledge of the real dimension of the squares. The effect of the perspective correction transformation is shown in Fig. 5.