Accessible remote sensing of bridge movement monitoring with UAV-based SfM photogrammetry and unsupervised machine learning

Bridge infrastructure in seismically active regions is highly vulnerable to damage from major seismic events. Post-disaster structural assessments are critical but are often hindered by traditional, contact-based methods that are labour-intensive, hazardous and slow. This study introduces a novel framework for rapid, remote assessment using a combination of Unmanned Aerial Vehicles (UAVs) and Structure from Motion (SfM) photogrammetry. By generating high-fidelity 3D point clouds, an unsupervised machine learning algorithm (RANSAC) is used to automatically extract key regions of interest (ROIs) without requiring labelled training data, enabling the precise measurement of post-seismic bridge movement. The analysis focuses on common failure modes: translational, rotational and settlement movements of supports. A systematic comparative analysis of three point-cloud-based change detection methods, M3C2, C2C and C2M, is conducted to identify the most suitable technique for different movement types. Experimental results validate the framework’s high accuracy and efficiency, presenting a viable, readily deployable solution to enhance the safety and resilience of critical infrastructure in disaster-prone areas.