An Image-Based System for Change Detection on Tunnel Linings

We present an automated system for detecting visual changes on tunnel linings. By registering new images to a three-dimensional tunnel surface model, recovered using Structure and Motion techniques, we are able to detect and localise changes accurately in order to assist visual inspection by a human expert. We formulate the problem of detecting changes probabilistically and exploit different feature maps and a novel geometric prior to achieve invariance to noise and nuisance sources such as parallax and lighting changes. System performance is assessed on a real data set labelled with ground-truth and collected with a prototype capture device. By localising, organising and classifying image data automatically, our system is a step towards higher frequency visual inspection at a reduced cost.