Aerial Object Detection

Aerial object detection studies how to find and classify objects in images captured from drones, aircraft, and satellites, where targets may be tiny, crowded, rotated, or partly obscured. Common approaches include YOLO-style detectors and DETR transformers, high-resolution feature pyramids and image slicing, oriented bounding boxes, and fusion of RGB with thermal or event-camera data. VisDrone and UAVDT test drone imagery, while DOTA and related benchmarks emphasize large, varied scenes and rotated objects; recent work also explores open-vocabulary recognition, weak supervision, and efficient deployment on edge devices. A central tension is improving small-object accuracy without excessive computation, while performance across weather, sensors, locations, and datasets remains difficult to guarantee.