Smartphone-Based Multimode Geomagnetic Matching/PDR Adaptive Fusion Positioning and Integrity Monitoring in a Variable Corridor Environment

Abstract

Extended Kalman filter (EKF) is commonly employed to integrate geomagnetic matching (GM) and pedestrian dead reckoning (PDR). However, an EKF with a constant stochastic measurement model using empirical or pretrained parameters restricts the applicability of geomagnetic/PDR fusion systems. To address this issue, we optimize the EKF-based magnetic/PDR fusion system from the perspectives of GM, stochastic measurement model, and localization error control. First, to improve the accuracy of GM, we optimize the observations of multimode GM (MMGM) by increasing the coverage of magnetic fingerprints. Second, we construct a parameter-free stochastic measurement model for the EKF framework by employing variance component estimation, PDR theoretical error, and relative displacements. Third, we propose a multilevel integrity monitoring (MLIM) algorithm for the state update, measurement update, and fusion state of the EKF to control positioning errors. Extensive experiments were conducted in a variable indoor corridor environment, and the results indicate that the proposed MMGM/PDR fusion system with the parameter-free EKF and an MLIM strategy exhibits comparable root mean square error (RMSE) for simple routes and a 22% lower RMSE for complex routes compared to the EKF utilizing a constant stochastic model. Furthermore, the proposed fusion system is error-tolerant to different walking speeds and device heterogeneity, showing a positioning error within 0.75 m for a test length of 210 m. The proposed system also outperforms several state-of-the-art magnetic/PDR fusion systems (using particle filter, adaptive EKF, deep learning-based method, etc.) comprehensively regarding positioning accuracy, workload, and computational complexity.