PSI - Issue 84

Francesco Allegrezza et al. / Procedia Structural Integrity 84 (2026) 81–88

84

The analysis of the synthetic videos reveals a systematic error affecting both subpixel algorithms when no blur is applied. Specifically, the error varies with the true subpixel position of the target, producing a sinusoidal pattern with a period of one pixel. The introduction of a moderate Gaussian blur significantly attenuates this effect, reducing both the amplitude of the oscillation and the overall estimation error. The results are shown in Fig.2.

a

b

Fig. 2. Systematic error found using quadratic interpolation (a) and the upsampling algorithm (b).

3.2. Case study 2: Laboratory tests To evaluate algorithm performance in a real but controlled scenario, a laboratory test with known and reproducible displacements was designed. A target was placed on the moving plane of an electromechanical tester (Controls UNIFRAME 250), which applied vertical translations measured by the displacement transducer integrated into the system. The target was held in position by a magnetic base secured to the moving plane. Video recordings were acquired using a monochrome global-shutter 2.3 megapixels industrial video camera (Teledyne FLIR Blackfly S BFS-U3-23S3M-C) equipped with a fixed 25 mm focal C-mount lens (TAMRON M112FM25). The camera was mounted on an aluminium tripod, connected via USB 3.0 to a laptop, and positioned at approximately 4 m from the target at the same height. The acquisition frequency was set to 50 frames per second (FPS). Lighting was controlled by a spotlight aligned with the shooting axis, and exposure settings were adjusted to obtain footage with zero gain. The target used was an aluminium composite panel with a white surface, coated with black matte film cut by laser according to the geometry of the AprilTag. The represented tag had an internal side length of 80 mm. The ROI selected for analysis had dimensions of 252×250 pixels, and the side of the target appeared in the image with a size of 128 pixels. The mm/pixel scale factor was determined following the procedure described in Micozzi et al. (2023), considering that the optical axis of the camera was oriented to be perpendicular to the plane of the target. To ensure consistency with the synthetic tests, two video sequences with identical motion patterns were acquired: one in focus and one with a slight blur deliberately introduced. The imposed motion was vertical and periodic, with a known amplitude of approximately 12 mm. Fig.3a illustrates the experimental setup and Fig.3b the selected ROI.