Comparative performance assessment of optical flow methods for vision-based structural identification using vibration data

This study presents a comparative performance assessment of four vision-based motion extraction approaches—Lucas–Kanade, Farnebäck, Horn–Schunck, and a phase-based method—for vibration identification and modal parameter estimation in a laboratory-scale single-story reinforced concrete (RC) frame subjected to controlled impact excitation. High-speed video recordings (216 fps) are processed through a custom-developed graphical user interface that standardizes region-of-interest (ROI) selection, pixel-to-length calibration, displacement extraction, and data export, enabling a fair algorithm-to-algorithm comparison under identical experimental conditions. Displacement time histories obtained from each method are evaluated in both the frequency domain using the Fast Fourier Transform (FFT) and in the modal domain via Operational Modal Analysis (OMA) implemented in the PyOMA_GUI environment. The comparison focuses on the low-amplitude vibration regime, where optical-flow robustness and noise sensitivity become critical. The results indicate that Lucas–Kanade and Farnebäck consistently identify the first natural frequency with strong repeatability across ROIs and yield modal stabilization patterns that agree well with OMA-based estimates. Horn–Schunck provides acceptable identification within the expected frequency band; however, its global smoothness constraint tends to attenuate true motion content and may introduce frequency shifts under low-amplitude responses. In contrast, the phase-based approach does not recover reliable modal parameters in the investigated configuration, exhibiting dominant components around 50 Hz rather than the structural mode near 26 Hz, suggesting susceptibility to experimental noise sources and insufficient motion content. Overall, the findings provide practical guidance for selecting optical-flow-based algorithms for vision-based structural identification and support their use as complementary tools in structural health monitoring applications.