Crack Detection and Localization for Concrete Structures Using UAV Incorporating Dual-Camera Systems With Heterogeneous Lenses

Abstract

Recent advances in unmanned aerial vehicles (UAVs) and optical imaging provide an alternative to traditional close-range human inspection of hard-to-reach areas in in-service bridges. Intelligent data acquisition and analysis techniques have been developed in recent studies to advance the level of automation and accuracy for such a UAV-based inspection approach. Nevertheless, simultaneous detection and 3D localization of microscale concrete cracks remains challenging due to the inherent trade-off between image resolution and field of view of optical cameras. This study proposes a UAV-based dual-camera inspection framework that integrates a telephoto camera for high-resolution crack imaging with a wide-angle camera for robust pose estimation and scene reconstruction. A two-stage crack segmentation network is developed by combining DeepLab v3+ with a point-based rendering (PointRend) refinement module to enhance crack boundary delineation in high-resolution imagery. In situ field experiments were conducted to validate the effectiveness of the proposed approach in crack delineation and localization, and the results demonstrate that an average crack localization error of 10.5 cm is achieved using a widely adopted DJI UAV platform and payload.