Residual Attention Network-Based Confidence Estimation Algorithm for Non-Holonomic Constraint in GNSS/INS Integrated Navigation System
Yimin Xiao,Haiyong Luo,Fang Zhao,Fan Wu,Xile Gao,Qu Wang,Lizhen Cui
DOI: https://doi.org/10.1109/tvt.2021.3113500
IF: 6.8
2021-11-01
IEEE Transactions on Vehicular Technology
Abstract:Nowadays, the availability of accurate vehicle position becomes more and more indispensable. The GNSS/INS (Global Navigation Satellite Systems/Inertial Navigation System) is currently the most widely-used integrated navigation scheme for land vehicles, which is capable of provide high-accuracy and continuous positioning results in the open-sky environments. However, under the GNSS-denied conditions, the existing GNSS/INS integrated system often fails to provide reliable positioning results due to various and nonlinear errors contained in the MEMS (Micro-Electro-Mechanical System) IMU (Inertial Measurement Unit) measurements. To improve the positioning accuracy during GNSS outage, deep learning has been introduced into the GNSS/INS integrated system in recent years. In this paper, we propose a residual attention network-based confidence (i.e., measurement noise covariance) estimation algorithm for non-holonomic constraint in GNSS/INS integrated navigation system, which adopts a residual attention network to dynamically estimate the noise covariance of the pseudo-observation (i.e., non-holonomic constraint) for optimal Kalman filtering (KF) fusion. To emphasize the more representative features with larger weights for accurate noise covariance estimation, we introduce an attention mechanism to automatically assign proper weights to the learned features according to their contributions. We evaluate our proposed method on three practical road datasets and compare it with other seven methods including the traditional KF, Pure INS, KF with three deep learning networks, K-means, and the Input-Delayed Neural Networks based method. Extensive experimental results demonstrate that our proposed RA-NHC bounds the errors associated with velocities and achieves reasonable accuracy improvement in position and velocity estimation.
telecommunications,engineering, electrical & electronic,transportation science & technology