Crack detection and 3D visualization of crack distribution for UAV-based bridge inspection using efficient approaches

Abstract

In order to improve the detection accuracy and efficiency of bridge crack detection models, while addressing the challenge of crack localization, this paper proposes an efficient Unmanned Aerial Vehicle (UAV)-based concrete bridge crack detection framework. The framework includes depth-based Regions of Interest (ROI) extraction, an improved YOLOv11 crack detection model, the SeaFormer lightweight crack segmentation model, an image quality assessment model, a pseudo-crack removal algorithm, the conversion of pixel values to actual values, and an efficient crack detection scheme. Comparative testing with mainstream models demonstrates the advantages of the proposed models in detection accuracy, localization accuracy, and lightweight design. Additionally, a multi-view 3D reconstruction scheme is proposed, offering lower memory and time requirements while improving performance. Combined with the aforementioned crack detection models, it achieves 3D reconstruction of bridge structures and visualization of the 3D distribution of cracks. In tests involving images of cracks from the piers of Zhongshan Bridge in Lanzhou, the crack identification accuracy reaches 93.2%, with an F1 score of 87.7% and a recall rate of 82.7%. The crack segmentation accuracy is 93.66%, and the Intersection over Union (IoU) is 90.17%. The results show that the proposed bridge crack detection framework delivers high lightweight performance and detection efficiency while maintaining high accuracy, making it more suitable for deployment on mobile devices such as UAVs for crack detection in bridges, towers, and other structures.