A learning-based method for structure vibration measurement

Due to the inconvenient use of sensors in structural health monitoring, vision-based, non-contact vibration measurement methods have been rapidly developed. However, most vision-based methods necessitate the installation of targets. Additionally, manually designed filters typically apply to a single scenario, exhibiting poor adaptability to diverse working conditions. This paper focuses on vision-based structural vibration measurement and introduces a novel learning-based method that combines motion magnification and edge detection technologies to address the issue of poor detection accuracy caused by small vibration amplitudes in full-field photographs. Vibration signals are extracted by detecting pixel changes at the edges of the structure. Furthermore, we use a learning-based motion magnification technique for structural modal identification. Experimental results from vibration detection tests on wind turbine towers demonstrate the effectiveness of our method. The system’s frequency detection error is less than 1.5% compared to accelerometers.