Vision-based structural displacement measurement under ambient-light changes via deep learning and digital image processing

Abstract

Structural displacement can provide critical information for structural safety assessment and maintenance. Although vision-based displacement measurement has significant advantages over traditional methods, it still faces challenges in long-term field applications owing to environmental uncertainties. This study proposes a novel displacement measurement method based on deep learning and digital image processing to mitigate the effects of ambient-light changes. The proposed method can automatically extract the calibration object in complex scenarios using You Only Look Once (YOLO) v5 and locate the calibration object under 24-h ambient-light changes precisely. Short- and long-term experiments were conducted in the laboratory to evaluate the performance of the method, and the short-term experimental results were compared with laser displacement sensor (LDS) data, which showed a maximum and minimum relative error of 0.4122% and 0.0024%. The long-term experimental results showed that the displacement responses were within ±1.0 mm. Hence, this method has good potential in structural displacement measurements.