Pixel-based operating modes from surveillance videos for structural vibration monitoring: A preliminary experimental study

Abstract

This paper presents a practical and cost effective method for vibration monitoring using multiple surveillance cameras distributed throughout (within) a building structure. Robust feature points in surveillance footage are identified and tracked frame-by-frame (with the Kanade-Lucas-Tomasi tracker) to extract the pixel motions. The relative pixel motions from multiple locations within the structure are analyzed in the frequency domain to estimate dynamic characteristics of the monitored structure (e.g., the dominant frequencies), as well as the pixel-based operating modes. These pixel-based modes importantly are in pixel units, eliminating the need to converted to engineering (physical) units. The method is verified through an experimental parametric study on a three-story lab-scale building structure (with three surveillance cameras) excited by broadband white noise signals. The accuracy and repeatability of the method are assessed to evaluate its application for structural health monitoring.