PSI - Issue 62

Fabio Micozzi et al. / Procedia Structural Integrity 62 (2024) 848–855 Author name / Structural Integrity Procedia 00 (2024) 000 – 000

850

3

mm square target was attached at midspan using an L-shaped metallic support. The video camera, at 12.10 m from the target and 60 cm below the deck, has optical axis is nearly parallel to the bridge axis. More details on the hardware and its technical specifications as well as on the target design can be found in Micozzi et al. (2023).

West first span deck

Accelerometer

West abutment

Displacement transducer

Target

Video camera

Monitored span

Eastbound dir.

Westbound dir.

Target South side

a

b

Deck cross section @ midspan

West pier

Fig. 1. (a) aerial view on the day of testing; (b) schematic view of the monitored span with indication of the adopted sensors.

A real-time video processing software was implemented in MATLAB, taking advantage of its computer vision toolbox as well as of the Generic Interface for Cameras (GenICam) protocol. The extraction of displacement time histories exploits a template matching algorithm in the version originally proposed by Guizar-Sicairos et al. (2008) and available as a MATLAB code. The algorithm is based on a cross-correlation peak matching, called upsampling cross-correlation (UCC), between a template selected by the user and each subsequent image of the video footage. The UCC template matching is intensity-based, i.e., the information of the image is mainly related to local intensity differences, therefore, better performs in case of high contrast targets, as the one adopted in this experimental campaign. The video processing workflow is the following. A portion of the image frame, called Region of Interest (ROI), is selected and within ROI a template is chosen. The motion of the template will be tracked inside the ROI. Digital noise in each image frame is reduced prior to the application of the template matching algorithm through a two dimensional Gaussian filtering Cross-correlation peak matching is performed to identify the displacement of the template within ROI in two steps: 1) a pixel-level rough search providing a preliminary estimation of the displacements with pixel resolution; 2) a subpixel fine search within a neighbourhood of the initial estimation achieving 1/  pixel resolution where  is the assigned upsampling factor (integer value), analytical details in Guizar Sicairos et al. (2008) as well as Feng and Feng (2021). The extracted displacements in two orthogonal directions in the plane perpendicular to optical axis are given as final output, without the need for storing the processed video frame (saving computer memory space). 2.3. Reference monitoring system A high-precision linear variable displacement transducer (Gestecno TSL-160) was installed at midspan using a stiff tripod anchored to the ground through steel bars inserted into the soil for about 1 m. This installation was made possible thanks to the reduced distance of about 1.8 m between the beam and the ground. A high-sensitivity piezoelectric accelerometer (PCB 393B31) measuring vertical accelerations was fastened at midspan, very close to the displacement transducer. Two piezoelectric accelerometers (PCB 393A03) were connected to the top of the tripod of the video camera (Figure 2) to measure accelerations in the plane perpendicular to the optical axis. The signals of the displacement transducer and accelerometers were acquired using a Dewesoft Krypton data logging system and its controlling software (Dewsoft X) with acquisition sampling frequency set to 1200 samples/s.

2.4. Measurements

Three measurements windows of 10 minutes each made during the monitoring campaign are analysed in this article. In the first one the video camera had acquisition frequency of 120 frames per second (FPS) and processing was made with upsampling factor  = 50, in the second acquisition  = 100 was adopted keeping 120 FPS, in the