TrafficDrone: A High-Resolution Multi-Altitude UAV Dataset for Traffic Object Detection

TrafficDrone is a high-resolution multi-altitude UAV dataset designed for traffic object detection and analysis. The dataset consists of 4K aerial images captured at different altitude levels ranging from 45 to 120 meters, providing significant scale variation for real-world scenarios. It includes annotated instances of nine traffic-related classes, such as cars, pedestrians, motorcycles, buses, and trucks, in YOLO format. The dataset contains 14,300 training images and 3,864 test images, with a total of over 700,000 annotated object instances. TrafficDrone is specifically designed to support research on small object detection, multi-scale learning, and UAV-based intelligent transportation systems. This dataset aims to provide a standardized benchmark for evaluating deep learning-based object detection models under varying altitude and scale conditions.