UAV-RDSD80

The UAV-RDSD80 dataset is a geo-referenced oblique UAV image dataset designed for pavement distress and traffic-sign detection on national and provincial trunk roads. The dataset was collected in Xi’an, Shaanxi Province, China, using a DJI Matrice 300 RTK UAV equipped with a Zenmuse P1 camera. Images were captured at an approximate altitude of 80 meters with an oblique viewing angle of approximately 45°, enabling simultaneous observation of road surfaces and roadside infrastructure.The dataset contains 905 high-resolution UAV images along with corresponding lane-based cropped images to reduce background redundancy. Each image is associated with structured metadata, including GPS coordinates, capture time, and camera parameters, enabling spatial analysis and reproducibility. Object annotations are provided in YOLO format, covering 13 categories of pavement markings, traffic signs, and pavement distresses, with a total of 23,645 annotated instances.To address challenges associated with large background regions and spatial correlation in UAV data, the dataset includes manually annotated lane-based regions of interest (ROI) and two complementary dataset splitting strategies. The segment-level split (Split-A) supports leakage-aware evaluation by separating spatially correlated samples, while the within-segment split (Split-B) provides balanced subsets for routine model development.The dataset is organized into a structured directory system, including raw images, cropped images, annotations, ROI definitions, metadata files, and predefined data splits. All components are linked through consistent filenames to ensure traceability and ease of use.UAV-RDSD80 provides a realistic and challenging benchmark for object detection in large-scale UAV imagery and supports research in intelligent transportation systems, infrastructure monitoring, and aerial visual perception.