Abstract Damage inspection on the undersides of bridges is an important and challenging part of routine bridge inspections. A method for 3D reconstruction and damage localization of bridge undersides based on close-range photography by unmanned aerial vehicle (UAV) and stereo vision combined with deep learning algorithms is proposed, the specific contributions include: (1) proposing a close-range photography method for acquiring high-resolution images from multiple perspectives of the bridge underside by UAVs, serving as the data source for damage analysis; (2) applying a deep learning-assisted segmentation method to optimize the multi-view geometry-based 3D reconstruction method, improving the efficiency of three-dimensional reconstruction, and defining the projection direction from the 3D reconstruction results to obtain ultra-high-resolution panoramic images of the bridge underside; (3) addressing the issue of detecting minor damages in large panoramic images by using a slice-assisted reasoning module and a lightweight convolutional YOLO v8 network to identify exposed steel bars corroded due to concrete damage in the panoramic images, and defining a coordinate system to localize the damages on the bridge underside. The proposed method was applied to damage detection and localization on the underside of a 160 m span main span of an in-service concrete bridge. The results demonstrate that the proposed method can quickly and accurately identify exposed steel bar corrosion on the bridge underside and output reports, proving the practicality of the proposed method.
Read full abstract