We present the OdomBeyondVision dataset, a comprehensive indoor odometry dataset containing the data from both emerging and traditional navigation sensors on diverse platforms. The dataset features:
-
The first indoor dataset to simultaneously cover the mmWave and LWIR thermal spectrum.
-
>10km sequences on UGV, UAV and handheld platforms in various indoor scenes and illumination conditions.
-
Extrinsic calibration parameters for all platforms and intrinsic parameters for all optic sensors, accompanied by several plug-and-play codes for developing multimodal odometry systems with our dataset.
The paper describing the dataset can found here.
To cite this dataset:
@misc{https://doi.org/10.48550/arxiv.2206.01589,
doi = {10.48550/ARXIV.2206.01589},
url = {https://arxiv.org/abs/2206.01589},
author = {Li, Peize and Cai, Kaiwen and Saputra, Muhamad Risqi U. and Dai, Zhuangzhuang and Lu, Chris Xiaoxuan and Markham, Andrew and Trigoni, Niki},
keywords = {Robotics (cs.RO), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {OdomBeyondVision: An Indoor Multi-modal Multi-platform Odometry Dataset Beyond the Visible Spectrum},
publisher = {arXiv},
year = {2022},
copyright = {Creative Commons Attribution 4.0 International}
}
The full dataset in the format of Rosbags can be downloaded from Google Drive.
Information for these sequences is recorded on Google Sheet.
Relevant ROS topics
/camera/color/image_raw: RGB image from D435/L515
/camera/depth/image_rect_raw: depth image from D435/L515
/flir_boson/image_raw: thermal image from FLIR Boson
/velodyne_points: point cloud data from Velodyne LiDAR
/imu/data: IMU data from Xsnes MTi
/odom: wheel odometry on UGV
/mmWaveDataHdl/RScan_left_scan: left radar on UGV
/mmWaveDataHdl/RScan_middle_scan: central radar on UGV or Handheld
/mmWaveDataHdl/RScan_right_scan: right radar on UGV
/vrpn_client_node/X500_pglite/pose: Vicon pose on UAV
/ti_mmwave/radar_scan_pcl_0: radar on UAV
Dataset toolkits are provided, to:
- Extract raw data from rosbags;
- Preprocess radar data fro milliEgo training;
- Package training data for milliEgo and DeepTIO.
Details can be seen here.
Calibration results for extrinsic parameters and intrinsic parameters are provided here.
Training, testing and evaluation for milliEgo and DeepTIO are provided here.
Pre-trained models for milliEgo and DeepTIO on all 3 platforms can be downloaded from Google Drive.
The data collected on AGV and handheld platforms are supported by Amazon Web Services via the Oxford-Singapore Human-Machine Collaboration Programme and EPSRC ACE-OPS (EP/S030832/1). The AGV and handheld sequences are collected at the University of Oxford.