Vision-based pedestrian-induced load tracking and estimation framework

초록

Pedestrian bridges are critical urban infrastructure but often experience serviceability problems due to pedestrian-induced vibrations. Vibration-based structural health monitoring (SHM) techniques, particularly system identification, are widely used to characterize bridge dynamic behavior. However, reliable system identification remains challenging because pedestrian-induced loads on bridges are difficult to measure and are often simplified as white noise, which may not accurately represent the actual excitation characteristics. Therefore, this paper proposes a vision-based pedestrian-induced load tracking and estimation framework that simultaneously monitors the location and magnitude of pedestrian-induced loads generated by tracked pedestrians in the form of time-varying vertical ground reaction forces (vGRFs) during bridge crossings. A hybrid tracking strategy that combines a low-cost tracking algorithm with a computationally lightweight person reidentification model is employed to track pedestrians via bounding boxes and unique IDs. Then, the pedestrian-induced loads of each tracked pedestrian, expressed as time-varying vGRFs, are estimated based on the vGRF-to-body-weight ratios at specific gait events derived from a large-scale published dataset. By integrating these two processes within a unified framework, both the location and magnitude of pedestrian-induced loads generated by each pedestrian can be monitored. The performance of the proposed framework is evaluated through laboratory-scale and on-site validations and compared with measurement data from sensors, demonstrating its feasibility and effectiveness in identifying pedestrian-induced loads on pedestrian bridges.

키워드