Automated Concrete Crack Inspection With Directional Lighting Platform

This paper presents the development and performance evaluation of a novel platform for visual concrete crack inspection. Concrete surfaces are imaged using directional lighting to support accurate crack detection, classification and segmentation. In addition to developing lab- and field- deployable hardware iterations, we outline customised convolutional neural networks and filters that leverage the directionally-lit data set. Crack classification and segmentation accuracies were both 10% higher than accuracies for standard imaging techniques with diffuse lighting, and crack widths of 0.1 mm were reliably detected and segmented. The major innovation described here is the combination of new hardware platforms for directional lighting, with a suite of algorithms that utilise the directionally-lit data set to improve crack detection and evaluation. This work demonstrates that directional lighting can improve the performance and robustness of automated concrete inspection. This could be key in supporting the efforts of asset managers as they seek to automate inspections of their ageing populations of concrete assets.