ABSTRACT
The monitoring and documentation of tunnel crack damages play a vital role in the tunnel asset management task of CERN, the European Laboratory for Particle Physics. To this end, this study presents and summarizes a method to identify the crack damaged tunnel sections and to digitalize crack information, using an integrated two-stage inspection approach that combines robot-mounted imaging, photogrammetry and deep learning. At the first stage, the deployed approach collects images with telemanipulated robotic systems, and individual images are stitched into a tunnel panorama to widen the inspection field of view. The subsequent crack maps reveal the crack distribution and highlight the severe crack-damaged tunnel chainages. At the second stage, high-resolution images are utilized to generate scaled orthomosaics, utilising Structure from Motion (SFM). Finally, deep learning methods are employed for pixel-level crack detection in high-resolution images for the extraction of crack location and quantification
