Magnetic Field-Enhanced Learning-Based Inertial Odometry for Indoor Pedestrian

Abstract

Pedestrian dead-reckoning (PDR) is a vital technique in pedestrian localization. Compared with traditional PDR, learning-based inertial odometry has the advantages of smaller position drift and is insensitive to pedestrian motion patterns. However, the heading drift of the trajectory is still the dominant error source for the position error drift in these methods. This study focuses on providing a pedestrian trajectory estimation method with low drift by properly fusing learned-based inertial odometry and magnetometer measurements under an indoor scenario containing significant magnetic field disturbance. The proposed method reduces the impact of magnetic field disturbance by adopting a long-term average magnetic vector, which is far more stable than using local magnetic vectors. Meanwhile, the proposed method can estimate the magnetometer bias online rather than depending on precalibrated magnetometer measurements. The test results show that the proposed method can obtain superior positioning performance using uncalibrated raw magnetometer data compared to other methods, even using calibrated magnetometer data. Simultaneously, this method achieves a balance between algorithm accuracy and efficiency.