Automated vision-based concrete crack measurement system

Maintaining and ensuring the safety of ageing infrastructures is becoming of uttermost importance. This is particularly critical in Europe, where several infrastructures like tunnels and bridges are approaching their end-of-life. Such a condition justifies the need for a prompt recognition of the deterioration status of these structures as well as the identification of strategies for effective inspection and monitoring operations. This paper focuses on cracks, and proposes a vision-based measurement system aimed at precisely and accurately detecting cracks on concrete surfaces and measuring their geometric features. The system mainly leverages RGB-D cameras to make the pixel to real world measurement units conversion possible (by assessing the camera to target relative distance and pose), and exploits an image processing pipeline consisting of the following steps: (a) an artificial intelligence-based semantic segmentation model for identifying cracks on the target structure; (b) a ridge detection algorithm for identifying the centre line of the cracks; (c) a geometric feature extraction algorithm to extract key metrics of the crack, like its width and length. Variance analysis of the crack mean width measurement shows a repeatability standard deviation of 0.01 mm and a reproducibility standard deviation of 0.02 mm, with a Type A expanded uncertainty of 0.004 mm. The system performance is discussed in the paper referring both to lab testing activities, targeted to the assessment of its metrological characteristics, and to a real field application, i.e. concrete inspection in a tunnel. This latter application is of particular interests as it is targeted to demonstrate the robustness of the approach in harsh environments (i.e., strong presence of dust, total absence of natural light) and on surfaces whose roughness generates important light gradients and shadows (e.g., sprayed concrete). • Automated system : Developed an automated vision system for concrete crack detection. • AI-driven segmentation : Accurately segments cracks using state-of-art AI algorithms. • Geometric feature analysis : Efficiently extracts and analyzes crack geometry for assessment. • Metrological compliance : Meets EU standards for accurate crack measurements. • Real-world application : Validated through testing, effective in tunnel inspections.