**SwiftBase: A dataset based on high-frequency visual measurement for **Visual-Inertial Localization in high-speed motion scenes

Abstract

​ This is a visual-inertia dataset for high-speed motion. The data collection platform uses two IDS U3-3041le-REV high-speed cameras and one Parker 3DMGQ7-GNSS/INS IMU. Keep the camera sampling frequency at 200Hz and compress the image in grayscale. The IMU sampling frequency is 200Hz.

​ The data platform and dataset are constructed and maintained by the Information Fusion Technology Laboratory of the School of Automation, Northwestern Polytechnical University, China. If you have any questions, please try contacting us:

Author: Zhenghao Zou, Chunhui Zhao, Xirui Kao, Jiangbo Liu, HaoChen Chai and Yang Lyu*

Contact information: zouzhenghao@nwpu.edu.cn

Paper: SwiftBase: A dataset based on high-frequency visual measurement for

Visual-Inertial Localization in high-speed motion scenes （IROS 2024）

---

Introduction

​ Using high-speed cameras can obtain more detailed process states, thereby improving the robustness and accuracy of visual SLAM. At the same time, high-frequency measured visual images can significantly improve feature tracking capabilities and alleviate image blur caused by high-speed motion. refer to:

@inproceedings{c9,
  title={Improving Localization Robustness in Monocular SLAM Using a High-Speed Camera},
  author={Gemeiner, Peter and Davison, Andrew J. and Vincze, Markus},
  booktitle={Robotics: Science and Systems IV},
  year={2008},
  organization={MIT Press}
}

Then we will introduce the experimental scenario and dataset. The research room scenario provides stable lighting and complex feature environments, and the data is collected using handheld devices; The laboratory has designed high-speed sports aimed at challenging extreme sports, using suspension ropes and rolling pulleys to achieve high-speed rotation and sliding. We offer only ROS bag formats and calibration result. To provide groud truth, a motion capture system, Optitrack, uses 360Hz for high-speed tracking. We performed some necessary processing on the data, including data timing check, data reduction and platform calibration.

We believe that we have provided some motion condition and designed some challenging problems for visual inertia SLAM, and hope to explore the practical application of VI-SLAM with relevant researchers in high-speed motion scene.

​ Now, this dataset has been open for public, you can get data from google drive or baidu drive. Thank you for using!

---

Examples

• platform

• sequence (low-comb)

• sequence (middle-comb)

• sequence (high-comb)

More details will open later!

---

Data Retrieve

Google Drive

https://drive.google.com/drive/folders/1lwLOps9IEREn25w_IaCFSfaC7x56qSB7?usp=drive_link

In groudtruth(GT), red point is start point and green point is end point.

Sequence	Duration（s）	Size（Mb）	Dist（m）	ROSbag	Translate	Rotation	GT
low_trans	90	708	23.218	low_trans	low	none
low_rot	113	798	12.485	low_rot	none	low
low_comb	122	845	22.333	low_comb	low	low
mid_trans	56.6	429	16.331	mid_trans	middle	none
mid_rot	106	707	12.552	mid_rot	none	middle
mid_trans_low_rot	59	356	7.603	mid_trans_low_rot	middle	low
low_trans_mid_rot	82	514	13.949	low_trans_mid_rot	low	middle
mid_trans_mid_rot	66	414	12.219	mid_trans_mid_rot	middle	middle
high_trans	79	550	11.185	high_trans	high	none
high_rot	111	775	16.618	high_rot	none	high
high_trans_low_rot	62	423	10.690	high_trans_low_rot	high	low
low_trans_high_rot	77	469	16.101	low_trans_high_rot	low	high
high_trans_mid_rot	56.9	315	16.020	high_trans_mid_rot	high	middle
mid_trans_high_rot	89	524	17.543	mid_trans_high_rot	middle	high
high_trans_high_rot	96	566	20.386	high_trans_high_rot	high	high
calibr	50	382	---	calibr	---	---	---
calibr_result	---	5.00	---	calibr_result	---	---	---

---

Baidu Netdisk

need time!

---

Reference

If you need academic citations, please follow the Bibtex format:

@inproceedings{SwiftBase,
  title={SwiftBase: A dataset based on high-frequency visual measurement for Visual-Inertial Localization in high-speed motion scenes},
  author={Zhenghao Zou, Chunhui Zhao, Xirui Kao, Jiangbo Liu, HaoChen Chai and Yang Lyu},
  booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={XX--XX},
  year={2024},
  organization={IEEE}
}