UAV-OBB

Dataset Description (≤3000 characters) UAV-OBB is an aerial urban vehicle dataset designed for rotation-aware object detection and smart-city traffic monitoring using oriented bounding boxes (OBBs). Many UAV/drone datasets annotate vehicles with axis-aligned rectangles that include unnecessary background and do not encode orientation; UAV-OBB provides tight rotated annotations to support accurate localization, orientation estimation, and downstream traffic analysis. The dataset contains 1,375 RGB images (JPEG) at 1920×1080 resolution captured from predominantly nadir-view UAV imagery over urban roads in Chongqing, Wuhan, and Beijing (China). It provides 35,615 oriented vehicle instances across six classes: bike, bus, car, other_vehicle, taxi, and truck. The data are split into 1,158 training images, 207 validation images, and 10 test images. Collection altitude was approximately 75–108 m under diverse real-world conditions, including morning, midday, evening, rain, and mist/light fog. Both wide field-of-view and zoom settings were used to introduce strong scale variation, ranging from small distant vehicles to large close-up buses and trucks. Annotations are provided in YOLOv8-OBB label format as plain text files (one per image). Each object is encoded by class id and a rotated rectangle represented by four corner vertices (x1,y1…x4,y4) in normalized image coordinates. All instances were manually annotated with rotation-capable tools and double-checked for quality and consistency. Occluded and truncated vehicles were included when the majority of the object was visible, reflecting realistic urban traffic scenes. To support practical evaluation beyond static images, UAV-OBB also includes supplementary MP4 video sequences: a short clip with sparsely annotated reference frames for human-verifiable temporal assessment, and a longer unannotated sequence for qualitative evaluation of stability and deployment behavior. UAV-OBB can be used for oriented detection, tracking, counting, density estimation, and traffic-flow analysis in urban UAV surveillance scenarios.