Target-free approach for vision-based structural system identification using consumer-grade cameras

TLDR

Structural health monitoring relies on system identification, which traditionally uses wired or wireless sensors to measure displacements, accelerations, and strains, but these approaches require limited sensor placement and additional equipment; recent computer‑vision methods employ high‑speed, high‑resolution cameras yet remain costly and target‑dependent. This study explores whether consumer‑grade cameras can perform target‑free structural system identification. The authors develop a target‑free displacement measurement pipeline that selects regions of interest, detects and tracks features, and removes outliers, and evaluate it by comparing the resulting model accuracy to that of a conventional wired system. The comparison demonstrates that the consumer‑camera approach achieves comparable model accuracy, highlighting its significant potential. © 2016 John Wiley & Sons, Ltd.

Abstract

Recent reports on America's infrastructure have emphasized the importance of structural health monitoring of civil infrastructures. System identification is a key component of many structural health monitoring strategies. Current system identification methods estimate models of a structure by measuring displacements, accelerations, and strains with wired or wireless sensors. However, these methods typically involve installation of a limited number of sensors at discrete locations and require additional data acquisition devices. To overcome these limitations, computer vision-based techniques have been introduced recently that employ high-speed and high-resolution cameras. Such cameras can be quite costly and require tedious installation of targets. This paper investigates the potential of using consumer-grade cameras for structural system identification without the need to install targets. The underlying methods for target-free displacement measurements are introduced, including region of interest selection, feature detection, point tracking, and outlier removal. A set of experiments are conducted to assess the efficacy of the proposed approach by comparing the accuracy of the identified model with one obtained using a conventional wired system. Careful comparison of the results demonstrates the significant potential of the proposed approach. Copyright © 2016 John Wiley & Sons, Ltd.