Air-to-Air Simulated Drone Dataset for AI-powered problems

This paper introduces the multi-view Air-to-Air Simulated Drone Dataset (A2A-SDD), a comprehensive simulated drone dataset captured using AirSim (c). The dataset encompasses diverse scenarios where one or two drones are pursued by one to three monitoring drones. It includes five types of drones, such as DJI models and a generic quadrotor model, recorded in various weather conditions and environments. Both loaded and unloaded drones are represented, and the dataset provides extensive annotations, including object detection and XYZ coordinates. The dataset offers potential applications in training deep learning-based models for counter-UAV measures such as localization and payload detection in single- and multi-view cases. Furthermore, preliminary experiments demonstrate the promising performance of trained networks on practical data, affirming the dataset's value in addressing real-world drone challenges using optical sensors. The synthetic dataset is publicly available on GitHub (https://github.com/CARG-uOttawa/Multiview-Air-to-Air-simulated-drone-dataset).