Mid-Air: A Multi-Modal Dataset for Extremely Low Altitude Drone Flights

Abstract

Flying a drone in unstructured environments with varying conditions is challenging. To help producing better algorithms, we present Mid-Air, a multi-purpose synthetic dataset for low altitude drone flights in unstructured environments. It contains synchronized data of multiple sensors for a total of 54 trajectories and more than 420k video frames simulated in various climate conditions. In this work, we motivate design choices, explain how the data was simulated, and present the content of the dataset. Finally, a benchmark for positioning and a benchmark for image generation tasks show how Mid-Air can be used to set up a standard evaluation method for assessing computer vision algorithms in terms of robustness and generalization. We illustrate this by providing a baseline for depth estimation and by comparing it with results obtained on an existing dataset. The Mid-Air dataset is publicly downloadable, with additional details on the data format and organization, at http://midair.ulg.ac.be.

Leveraging Synthetic Data in Object Detection on Unmanned Aerial Vehicles

“…The need for data sets caused many researchers to focus on synthetic data. Mid-Air [15] presents a synthetic data set for unstructured environments captured with the Unreal Engine [22] in combination with AirSim [20]. They laboriously built a world with landscape and streets for drone navigation.…”

“…Few works consider the generation of synthetic data captured from UAVs. However, these only focus on the capturing process of the sensors and are lacking in world, object and physics details, or do not feature them at all [12], [15].…”

The GRIFFIN Perception Dataset: Bridging the Gap Between Flapping-Wing Flight and Robotic Perception

“…In recent years, a wide variety of drone datasets have been introduced for robotic perception tasks, see e.g., [6]- [12]. The work in [6] presents Mid-Air, a large synthetic dataset of quadcopters flying in unstructured environments that includes data from different scenarios and climate conditions using the Airsim simulator.…”

“…In recent years, a wide variety of drone datasets have been introduced for robotic perception tasks, see e.g., [6]- [12]. The work in [6] presents Mid-Air, a large synthetic dataset of quadcopters flying in unstructured environments that includes data from different scenarios and climate conditions using the Airsim simulator. The work in [8] provides a dataset including real and photo-realistic synthetic data to evaluate place recognition methods with respect to viewpoint tolerance.…”

Towards Real-Time Monocular Depth Estimation for Robotics: A Survey

“…This results in the perspective and angle of view of the captured images being different from the UAV view. Therefore, this may raise concerns about the generalization potential of the trained CNNs to the application of obstacle avoidance for UAVs [60].…”