Virtual sensor grids for full-field vibration measurement via superpixel segmentation and phase-based optical flow

This study presents a novel vision-based approach for high-resolution structural vibration measurement by integrating phase-based full-field motion estimation with virtual sensor grids defined through vibration-guided superpixel segmentation. To overcome the noise sensitivity and instability associated with conventional pixel-level methods, we introduce virtual sensor grids, which cluster spatially and dynamically coherent regions within video data. This strategy enables interpretable and noise-resilient vibration analysis. Within each virtual sensor unit, we perform confidence-weighted spatial aggregation based on a pixel-wise confidence metric derived from phase nonlinearity, resulting in robust and accurate displacement estimation. Experimental validation on an air compressor system demonstrates that the proposed method achieves displacement accuracy comparable to that of a laser Doppler vibrometer (LDV) and facilitates effective structural damage detection without the need for speckle patterns or physical markers. These results confirm the method’s suitability for structural diagnostics, particularly in environments where sensor or marker placement is challenging or where long-term, non-intrusive monitoring is required.