Automatic complex concrete crack detection and quantification based on point clouds and deep learning

The inspection of cracks on structural surface is an essential component of structural condition assessment. Recent advancements in image-based techniques aim to automate this process and reduce the associated costs. However, these methods typically focus on localized planar cracks, neglecting the more severe and complex multi-planar cracks. Moreover, accurate global positioning and quantification of these complex cracks pose challenges due to image distortions. To tackle these challenges, this paper proposes a point cloud-based framework for the automatic detection and quantification of complex cracks on concrete surfaces. The proposed framework involves generating a semantic point cloud for crack positioning and quantification by integrating a pixel-level crack segmentation network with 3D reconstruction algorithms. With the establishment of the semantic point clouds, a crack skeletonization based crack length quantification method and a crack width quantification algorithm are proposed to facilitate the automated and effective crack quantification. Experimental evaluations are conducted on a test beam with five characteristic flexural cracks. The results demonstrate the superior performance of the proposed framework over conventional methods with a mean relative error of merely 1.3 % in length quantification and a median of absolute errors less than 0.15 mm in crack width measurements. The efficacy and precision of the proposed framework enhance the potential applications of vision techniques for crack inspection. • Propose a point cloud-based framework for automated concrete crack detection and quantification. • Combine crack segmentation and 3D reconstruction to generate semantic point clouds. • Develop a skeletonization-based method for efficient crack quantification. • Enable precise crack localization and high-accuracy length and width measurements. • The effectiveness and accuracy of the proposed approach are demonstrated experimentally.