Core Concepts
The LuViRA dataset provides synchronized data from vision, 5G radio, and audio sensors captured in a controlled indoor environment, enabling research on sensor fusion for accurate localization.
Abstract
The LuViRA dataset is a comprehensive, publicly available dataset that includes synchronized data from vision, 5G radio, and audio sensors, as well as accurate 6DOF ground truth, captured in an indoor environment. The dataset consists of 89 trajectories recorded using a mobile robot equipped with a camera, antenna, and speaker, along with 12 microphones placed in the environment.
The key highlights of the dataset include:
Vision data: The dataset includes RGB, depth, and inertial measurement unit (IMU) data captured at 15-30 fps. The camera is calibrated intrinsically and extrinsically.
Radio data: The dataset includes channel response measurements between a 5G massive MIMO testbed and a user equipment (UE) mounted on the robot, captured at 100 Hz.
Audio data: The dataset includes audio recordings from 12 microphones placed in the environment, with a sampling rate of up to 96 kHz. The microphones are synchronized using a reference microphone on the robot.
Ground truth: The dataset includes 6DOF ground truth position and orientation data captured by a high-accuracy motion capture system, with an error of less than 0.5 mm.
Calibration and synchronization: The sensors are carefully calibrated and synchronized to ensure accurate data alignment, enabling the use of the dataset for sensor fusion research.
The dataset is divided into "grid" and "random" trajectories, with the grid data providing dense spatial sampling for radio-based localization algorithms, and the random data capturing more dynamic environments. The dataset is validated using state-of-the-art localization algorithms for each sensor modality, providing a baseline for future research.
The LuViRA dataset aims to enable research on sensor fusion for accurate indoor localization, as well as other applications such as channel estimation and image classification. The dataset is publicly available and can be accessed at https://github.com/ilaydayaman/LuViRA_Dataset.
Stats
The vision system captures RGB, depth, and IMU data at 15-30 fps.
The radio system captures channel response measurements between a 5G massive MIMO testbed and a user equipment at 100 Hz.
The audio system records 12 microphone channels at up to 96 kHz sampling rate.
The ground truth system provides 6DOF position and orientation data at 100 Hz with an error of less than 0.5 mm.
Quotes
"To build an autonomous smart factory, one of the most critical challenges is performing accurate localization and monitoring of autonomous service robots in real-time for such an indoor environment, e.g., a service robot in a factory needs to localize itself within centimeter-level accuracy to perform tasks such as lifting and placing objects."
"Recent works [10], [11] indicate that these sensors can be used jointly to complement each other and enhance the overall performance. However, to develop, evaluate, and compare algorithms that fuse these sensors, a public dataset that includes simultaneous and synchronized sensor readings from each sensor in the same environment is required."