Robust and versatile vision-based dynamic displacement monitoring of natural feature targets in large-scale structures

Dynamic displacement response is an essential indicator for assessing structural state and performance. Vision-based structural displacement monitoring is considered as a promising approach. However, the current vision-based methods usually only focus on certain application scenarios. This study introduces a Sparse Bayesian Learning-based (SBL) algorithm to enhance robustness, accuracy, and computational efficiency in target tracking. Furthermore, a robust and versatile Vision-based Dynamic Displacement Monitoring System (VDDMS) was developed, capable of monitoring displacements of varying application scenarios. The robustness of the proposed algorithm under changing illumination conditions is validated through a specially designed indoor experiment. The feasibility of field application of VDDMS is confirmed through an outdoor shear wall shaking table test. Furthermore, a large-scale bridge shaking table test is conducted to evaluate the reliability and versatility of VDDMS in monitoring natural feature targets on large structures subjected to different seismic excitations. The root mean square error, when compared to laser displacement sensors, ranges from 0.2% to 2.9% of the peak-to-peak displacement. Additionally, VDDMS accurately identifies multi-order frequencies in bridge structures. The study investigates the influence of initial template selection on accuracy, highlighting the significance of distinctive texture features. Moreover, two error evaluation schemes are proposed to quickly assess the reliability of vision-based displacement sensing technologies in various application scenarios.