EXPERIMENTAL VALIDATION OF FREQUENCY AND MODAL SHAPE OF TWO BUILDINGS UNDER AMBIENTAL DISPLACEMENTS USING DRONE VIDEOS WITHOUT FIDUCIAL MARKERS.

In civil engineering, accelerometers are widely used for vibration-based Structural Health Monitoring (SHM) analysis. While both wired and wireless accelerometers offer high sensitivity and resolution, wired sensors are constrained by installation requirements, and wireless sensors need individual power sources, making them susceptible to interference and data transmission issues. Additionally, sensor installation is time-consuming and can lead to damage or calibration problems. To address these challenges, camera-based vibration measurement has gained attraction. Cameras, whether can be installed on fixed supports (stationary) or drone-mounted, have been used to measure dynamic and static displacements in structures such as bridges, antennas, and suspension cables. Many studies rely on fiducial markers for tracking, but marker placement hinders convenience. Targetless measurement, which only requires the camera, has emerged as a practical solution, particularly for capturing large displacements. In this study, we filmed two buildings—a nine-story and a 54-story structure—using a commercial drone equipped with an integrated camera (20MP camera, 4K videos with 30fps for 5 min). The recordings were made without fiducial markers under ambient vibrations. By employing computer vision techniques, we tracked the buildings' distinctive features over time in the video footage. We separated the resulting time series into low and high-frequency components using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). We expected the low-frequency component to represent the drone's displacement, while the high-frequency component contained the building's displacement under ambient excitation and noise. We further analyzed the high-frequency time series using a robust algorithm that consider the large number of observed displacements in several regions of interest. We use Covariance-based Stochastic Subspace Identification (SSI-COV) to extract frequency, modal shape, and damping ratio. Comparing these results with data from accelerometers on the structures, we found accurate frequency and modal shape identification, with slight discrepancies in damping estimation. This approach shows promise for enhancing the accuracy of frequency, damping, and modal shape analysis in future SHM investigations, particularly when capturing substantial structural displacements.