Investigation of Data-Driven Calibration and Intelligent Filtering Technique for Robust Foot-Mounted Pedestrian Localization

In inertial measurement unit (IMU)/magnetometer integrated pedestrian positioning systems, the absolute heading from the magnetometer is severely degraded by pervasive magnetic interference, introducing uncertainty into the integrated solution and reducing overall positioning accuracy. Conventional ellipsoid-fitting methods for magnetometer calibration typically rely on minimizing the algebraic distance; however, they perform poorly under highly inhomogeneous data distributions. Moreover, the unknown relationship between heading-error statistics and magnetometer orientation in 3-D space can significantly degrade system performance. Addressing these challenges, this study proposes a data-driven progressive ellipsoid-fitting method to mitigate deterministic magnetic interference. In parallel, an adaptive extended Kalman filter (EKF) enhanced by a multilayer perceptron (MLP) models the statistical characteristics of heading errors in various spatial orientations. By dynamically adjusting the observation noise matrix in real time, the system effectively suppresses stochastic magnetic interference. This approach is especially suitable for foot-mounted pedestrian positioning, which typically exhibits limited motion degrees of freedom and only small rotational movements during normal use. Furthermore, it automatically estimates the observation noise level of IMU/magnetometer integrated positioning systems without any prior knowledge. The experimental results confirm that the proposed method delivers notable effectiveness in real-world scenarios.