A Low-Altitude UAV Dataset Based on Visual–Inertial Navigation

Abstract

As unmanned aerial vehicles (UAVs) are increasingly utilized in operational domains ranging from 50 to 500 m, there is a growing need for robust and reliable autonomous positioning, particularly in scenarios where satellite signals are unavailable. To address this challenge, we present a set of visual–inertial datasets collected by low-altitude UAVs. The dataset includes stereo cameras, an inertial measurement unit (IMU), a laser ranging sensor, and RTK-based ground truth. All these sensors have been calibrated and synchronized. The dataset consists of 28 sequences captured across diverse scenarios. To the best of our knowledge, this is the first UAV visual navigation dataset to encompass such a wide range of flight altitudes. Furthermore, we have evaluated the dataset using state-of-the-art visual–inertial simultaneous localization and mapping (SLAM) algorithms. The results demonstrate that existing solutions perform inadequately in certain situations. For the benefit of the research community, we make this dataset publicly accessible. The webpage of our dataset is https://github.com/boomzoeycy/A-Low-Altitude-UAV-Dataset-based-on-Visual-Inertial-Navigation .