FDIO: Extended Kalman Filter-Aided Deep Inertial Odometry

Yingying Wang1; Hu Cheng1; Max Q.-H. Meng2

10.1109/ICARM58088.2023.10218871

2023

979-8-3503-0018-5

2023 International Conference on Advanced Robotics and Mechatronics (ICARM)

482-487

8-10 July 2023

Sanya, China

Smartphone-based deep inertial odometry has recently gained great research interest, which utilizes the inertial measurement unit (IMU) and deep learning technique for relative states estimate. In this paper, we propose FDIO: a Filter-aided Deep Inertial Odometry, which utilizes an extended Kalman filter (EKF) for orientation tracking following by a learning module for position estimation. Our inertial odometry requires only the inertial signals without relying on external device orientation information. In the position learning module, this paper proposes a novel representation of the pedestrians' velocity and a robust loss for regression. The proposed FDIO is validated by using the public RoNIN dataset. Experimental results show that our model outperforms state-of-the-art deep inertial odometry architectures.

Deep learning Pedestrians Mechatronics Measurement units Estimation Inertial navigation Information filters

其他

IEEE

1.Department of Electronic Engineering, Robotics, Perception and Artificial Intelligence Lab, The Chinese University of Hong Kong, Hong Kong SAR, China
2.Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen, China

Yingying Wang,Hu Cheng,Max Q.-H. Meng. FDIO: Extended Kalman Filter-Aided Deep Inertial Odometry[C],2023:482-487.