ABSTRACT
Effective asset management of underground infrastructure requires timely and detailed visual inspections for condition monitoring. To date, the common approach for visual inspections is heavily manual and can be slow, expensive, and subjective to the engineer’s experience. The present paper describes the end-to-end development of an automated visual inspection workflow. The proposed two-stage pipeline applies the state-of-the-art Deep Learning (DL) algorithms for object detection and tracking to identify structural defects in tunnels. Then, via a novel computer vision approach, it detects and tracks natural features for a precise sensor-less in-tunnel defect mapping. The output is a DL-powered digital twin of the inspected infrastructure. It drastically reduces manual input for repetitive tasks and focuses on the employment of highly trained engineers for validation purposes only. This means that engineering knowledge is more effectively spent. These results will impact the construction industry’s approach to visual inspections shifting it towards automated strategies.
