Enhancing vision-based structural monitoring through data fusion with sparse accelerometer measurements

Dynamic monitoring plays a vital role in civil engineering by enabling accurate assessment and safeguarding of structural performance under operational and extreme loads. Traditional accelerometer-based systems, while reliable and time-detailed, require extensive wiring, high costs, and face challenges in deriving displacement data due to integration errors, especially in the case of residual displacements or non-zero-mean displacements. Vision-based techniques offer a non-contact alternative, capable of wide-area displacement measurements with minimal installation, but their standalone use is limited by environmental sensitivity and accuracy constraints. This paper proposes a data fusion framework that integrates sparse accelerometer data with vision-based displacement measurements to provide a cost-effective, minimally invasive solution for dynamic structural monitoring. Employing only a limited number of accelerometers, the method reconstructs acceleration responses at un-instrumented but visually monitored locations through modal-based virtual sensing, enabling comprehensive spatial and temporal monitoring without dense sensor networks. The fusion of reconstructed accelerations and vision-derived displacements enhances measurement reliability and robustness. The approach is validated via experimental tests on a pedestrian bridge in Modena (Italy), demonstrating strong agreement with reference data and confirming its suitability for practical deployment in real-world infrastructure monitoring.