Abstract: Computer vision-based structural vibration identification methods have achieved significant research results and remarkable progress in engineering applications. This paper first summarizes and reviews traditional computer vision-based methods for structural vibration identification, highlighting the current research limitations in recognizing small-amplitude structural vibrations. To address these limitations, the paper systematically introduces corresponding improved methods in three progressive aspects: automatic camera vibration compensation (Section 2), monocular camera-based small-amplitude two-dimensional (2D) vibration identification (Sections 3 and 4), and multi-view three-dimensional (3D) small-amplitude vibration identification (Section 5). Specifically, in terms of camera automatic vibration compensation, related methods utilize the maximum likelihood estimation of Bayesian models to detect pairs of feature points from stationary background regions, thereby estimating a joint homography matrix that only contains camera vibration information, which lays the foundation for subsequent structural vibration identification. Based on this, for monocular small-amplitude 2D vibration identification, relevant methods employ multi-frequency phase retrieval models to detect the phase information under noise and eliminate phase wrapping effects. In addition, a compressed sensing-based sparse enhancement technique is introduced to avoid phase constraint limits and mitigate ill-posed estimation impacts. Furthermore, to tackle a 3D vibration identification, related methods use weighted phase unwrapping to address phase wrapping and ill-posed estimation issues, and then apply reweighted minimization in multi-view estimation to more thoroughly eliminate outliers. However, the research on structural small-amplitude vibration monitoring based on computer vision is still in its early stages. Future researches should focus on the vibration identification under complex environments, on the nonlinear and large-amplitude vibration identification, on the long-term monitoring and full life-cycle evaluation, on the optimization of 3D vibration identification methods, as well as on the integration and engineering application of algorithms to reduce errors caused by various factors and to improve efficiency and reliability in structural health monitoring.