Video-Based Vibration Measurement for Large Structure: A Spatiotemporal Disturbance-Adaptive Morphological Component Analysis

Recently, phase-based motion estimation method is able to extract the full-field vibration of large structure from video, which has attracted widely attention. However, it suffers from diverse disturbances in the realistic measurement, such as periodic texture pattern from the surface and shake caused by unstable tripod. To address this issue, a spatiotemporal disturbance-adaptive morphological component analysis (DAMCA) is proposed in this article. This method focuses on separating texture component from structure component in each video frame and separating shake component from vibration component in the global motion signal extracted from video. First, it constructs a spatiotemporal model to remove the disturbances, respectively, from the space domain and time domain of the video. Subsequently, it incorporates morphological component analysis (MCA) into this model, aiming to obtain the optimal separation results in the space domain and time domain by finding the sparsest representations of different components. Finally, to adapt to various texture surfaces, MCA is developed in the space domain by exploiting the low-rank structure of the texture's sparse representation. Similarly, to adapt to various shake, MCA is also developed in the time domain. Simulated and actual test demonstrate the effectiveness and accuracy of the proposed method. Spatiotemporal DAMCA has valuable potential to acquire video-based vibration measurement for large structure with the disturbances involving texture surface and tripod shake.