Use of a roving computer vision system to compare anomaly detection techniques for health monitoring of bridges

Displacement measurements can provide valuable insights into structural conditions and in-service behaviour of bridges under operational and environmental loadings. Computer vision systems have been validated as a means of displacement estimation; the research developed here is intended to form the basis of a real-time damage detection system. This paper demonstrates a solution for detecting damage to a bridge from displacement measurements using a roving vision sensor-based approach. Displacements are measured using a synchronised multi-camera vision-based measurement system. The performance of the system is evaluated in a series of controlled laboratory tests. For damage detection, five unsupervised anomaly detection techniques: Autoencoder, K-Nearest Neighbours, Kernel Density, Local Outlier Factor and Isolation Forest, are compared. The results obtained for damage detection and localisation are promising, with an f1-Score of 0.96–0.97 obtained across various analysis scenarios. The approaches proposed in this research provide a means of detecting changes to bridges using low-cost technologies requiring minimal sensor installation and reducing sources of error and allowing for rating of bridge structures.