Three-Dimensional Crack Quantification Using Fused LiDAR Data and Optical Images

Abstract

Cracking is a common issue in concrete structures, yet current assessment methods are not able to adequately quantify cracks in 3D spatial dimensions. This study proposes a simplified approach utilizing optical cameras and light detection and ranging (LiDAR) technology, combining image and point cloud data for efficient and accurate crack evaluation. A Local Feature Transformer (LoFTR)-based image registration method is introduced to align low-texture point cloud intensity data with optical images, enabling pixel-level resolution mapping of the optical images. Contour line feature enhancement is employed to improve the quality and quantity of matching points. Furthermore, a crack detection and quantification framework is proposed, which comprises the YOLOv8-CBAM-seg model for crack detection and an improved orthographic projection (OP) method for crack width quantification. The proposed model achieves a segmentation accuracy of 96.7% and an inference speed of 53 frames per second, enabling precise and rapid crack identification and segmentation. By integrating the spatial information from LiDAR data with the pixel-level details provided by YOLOv8-CBAM-seg and the improved OP method, 3D spatial crack quantification is realized. An experiment was conducted to detect cracks in a concrete bridge pier, achieving millimeter-level accuracy in crack width measurement, with average relative errors of 2.48% for planar surfaces and 11.74% for curved ones. The results validate the feasibility and practical applicability of the proposed approach for routine maintenance and safety assessment of concrete structures with a cost-effective and efficient solution.