SuperVINS: A Real-Time Visual-Inertial SLAM Framework for Challenging Imaging Conditions

IEEE Journal in submission
Hongkun Luo¹ Yang Liu² Chi Guo^3* Zengke Li⁴ Weiwei Song^5
1 Wuhan University ² China University of Mining and Technology
^* represents the corresponding author.   

[Paper]      [Code]    

SuperVINS framework

SuperVINS is a visual-inertial SLAM framework that integrates deep learning features based on VINS-Fusion. It has the characteristics of high accuracy and high robustness.

Abstract

There is an increasing emphasis on achieving high accuracy and robustness in SLAM systems. Traditional visual-inertial SLAM system often struggles with stability under low-light or motion-blur conditions, leading to potential lost of trajectory tracking. High accuracy and robustness are essential for the long-term and stable positioning capabilities of SLAM systems. Addressing the challenges of enhancing robustness and accuracy in visual-inertial SLAM, this paper propose SuperVINS, a real-time visual-inertial SLAM framework designed for for challenging imaging conditions. Developed as an enhancement of VINS-Fusion, SuperVINS integrates the deep learning neural network model SuperPoint for feature point extraction and loop closure detection. At the same time, a deep learning neural network LightGlue model for associating feature points is integrated in front-end feature point matching. A feature points matching optimization strategy based on the RANSAC algorithm is proposed. The system is capable of dynamically adjusting thresholds in response to environmental changing, thereby balancing computational cost and positioning accuracy. Additionally, it allows for flexible training of specific SuperPoint bag of words tailored for loop closure detection in particular environments.The system enables real-time positioning and mapping without the need for preprocessing. Experimental validation on the well-known EuRoC dataset demonstrates that SuperVINS outperforms in accuracy and robustness across most challenging sequences. This paper analyzes the advantages of SuperVINS in terms of accuracy, real-time performance, and robustness.

Experimental Results

BibTex

@article{supervins,
  title={SuperVINS: A visual-inertial SLAM framework integrated deep learning features},
  author={Hongkun Luo and Yang Liu and Chi Guo and Zengke Li},
  journal={arXiv preprint arXiv:2407.21348},
  year={2024}
}

Acknowledgements: We borrow this template from A-tale-of-two-features.