A dynamic initialization method for visual inertial odometry based on deep learning

Qinglin Liu; Shuai Jia; Hao Li; Yong Zhang

doi:10.1117/12.3045786

13 December 2024 A dynamic initialization method for visual inertial odometry based on deep learning

Qinglin Liu, Shuai Jia, Hao Li, Yong Zhang

Proceedings Volume 13492, AOPC 2024: Laser Technology and Applications; 1349205 (2024) https://doi.org/10.1117/12.3045786
Event: Applied Optics and Photonics China 2024 (AOPC2024), 2024, Beijing, China

Abstract

Visual inertial odometry (VIO) requires accurate initial parameters before navigation, such as the system's initial pose, scale information, and inertial measurement unit (IMU) biases. Consequently, rapid and precise initialization is crucial for ensuring the smooth progress of subsequent navigation. However, when the system is in an environment lacking visual features, the system without measurement constraints will rapidly diverge. During this period, due to the unknown motion state, static initialization cannot be performed. To ensure that the system can navigate through this period as smoothly as possible and restart VIO, this paper proposes a dynamic initialization method for visual inertial odometry based on deep learning. This approach relies solely on inertial data, using a deep learning network to learn attitude errors and uncertainties, which are then utilized as measurement values and combined with an extended Kalman filter (EKF) to correct the system state. Experimental results on a public dataset show that the proposed method enables rapid dynamic initialization under short-term visual measurement absence, and effectively improves the system's attitude accuracy. Compared to the method of traditional inertial navigation solving after gyroscope calibration, the heading error of our proposed method is reduced by 13.38%.

(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Qinglin Liu, Shuai Jia, Hao Li, and Yong Zhang "A dynamic initialization method for visual inertial odometry based on deep learning", Proc. SPIE 13492, AOPC 2024: Laser Technology and Applications, 1349205 (13 December 2024); https://doi.org/10.1117/12.3045786

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available