A data set of aerial imagery from robotics simulator for map-based localization systems benchmark

TL;DR

This paper introduces AIR, a large public dataset of aerial imagery from a robotics simulator, designed to benchmark map-based localization, visual odometry, and SLAM algorithms for high-altitude UAV flights.

Contribution

It provides the largest publicly available aerial imagery dataset from a robotics simulator for high-altitude UAV localization research.

Findings

Largest available public dataset with downward facing camera imagery

Contains over 100,000 images from urban and forest environments

Supports development and benchmarking of high-altitude UAV localization algorithms

Abstract

Purpose: This paper presents a new dataset of Aerial Imagery from Robotics simulator (abbr. AIR). AIR dataset aims to provide a starting point for localization system development and to become a typical benchmark for accuracy comparison of map-based localization algorithms, visual odometry, and SLAM for high altitude flights. Design/methodology/approach: The presented dataset contains over 100 thousand aerial images captured from Gazebo robotics simulator using orthophoto maps as a ground plane. Flights with 3 different trajectories are performed on maps from urban and forest environment at different altitudes, totaling over 33 kilometers of flight distance. Findings: The review of previous researches shows, that the presented dataset is the largest currently available public dataset with downward facing camera imagery. Originality/value: This paper presents the problem of missing publicly available datasets for high altitude (100--3000 meters) UAV flights, the current state-of-the-art researches performed to develop map-based localization system for UAVs, depend on real-life test flights and custom simulated datasets for accuracy evaluation of the algorithms. The presented new dataset solves this problem and aims to help the researchers to improve and benchmark new algorithms for high-altitude flights.