Non-stationary vision sensing for time-frequency analysis in vehicle-bridge interaction system

Global monitoring of structures is vital for assessing their structural integrity, especially with the impact of moving vehicles on railroad bridges. This necessitates simultaneous monitoring of both systems to understand interaction dynamics comprehensively. In vibration-based Structural Health Monitoring fields, demands for directly obtaining displacement responses increase, leading to non-contact sensing adoption. Computer Vision (CV)-based methods, using feature tracking techniques for displacement measurements, have become practical alternatives. The proposed approach utilizes Poor Feature Points, offering a global view and overcoming spatial resolution limitations. Addressing challenges related to camera ego-motion in large-scale monitoring, strategies for re-assigning regions of interest based on feature quality are introduced, and camera ego-motion is compensated by calibrating feature points. The You Only Look Once algorithm is used for vehicle wheel detection, localizing contact points to examine Vehicle-Bridge Interaction dynamics. A laboratory-scale experiment validation confirms the feasibility of global monitoring with vision sensors, especially in interpreting VBI dynamics.