Crack detection and quantification for concrete structures using UAV and transformer

Crack detection is of significant importance for concrete structural inspection. Unmanned aerial vehicle (UAV)-based crack detection systems abound, but simply quantifying cracks remains a challenge. This article aims to develop a UAV-based approach to achieve accurate detection and quantification of concrete cracks without reference markers. An improved calibration method is developed that establishes the full-field scale of UAV gimbal camera, and the scale factor can thus be conveniently indexed at various measurement attitudes without recalibrating. Additionally, an independent boundary refinement transformer (IBR-Former) is proposed for crack segmentation from UAV-captured images, wherein the IBR scheme can further refine crack boundary in a model-agnostic manner. The proposed framework can quantify the cracks with widths less than 0.2 mm, which meets the requirements of civil engineering codes. The on-site operational feasibility and practicability in engineering practice of the proposed framework are demonstrated by UAV field detection of multiple buildings and bridges.