Automated concrete crack evaluation using stereo vision with two different focal lengths

Surface cracks in concrete structures are an important indicator of the soundness of a structure. Stereo vision, consisting of two identical cameras, has been suggested to quantify crack characteristics using derived depth information. However, because high measurement resolution is required, zoom lenses are often used, making simultaneously crack localization and characterization difficult. This study presents a framework for the use of stereo vision employing one wide-angle lens and one telephoto lens, enabling accurate crack quantification as well as efficient 3D reconstruction. Furthermore, a robust depth estimation strategy is proposed for planar surfaces, such as are found in most concrete bridges. The performance of the proposed approach is field validated using an in-service concrete bridge. The 3D reconstruction model generated by a set of wide-angle images, including crack information extracted from the telephoto images using deep learning, can enable the improved inspection of concrete structures.