Smartphone-Based Multi-Mode Geomagnetic Matching/PDR Adaptive Fusion Positioning and Integrity Monitoring in a Variable Corridor Environment
Kefan Shao,Zengke Li,Meng Sun,Zhisheng Zhao,Qiang Guo
DOI: https://doi.org/10.1109/jiot.2024.3496522
IF: 10.6
2024-01-01
IEEE Internet of Things Journal
Abstract:Extended Kalman filter (EKF) is commonly employed to integrate geomagnetic matching and pedestrian dead reckoning (PDR). However, an EKF with a constant stochastic measurement model using empirical or pre-trained 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 geomagnetic matching, stochastic measurement model, and localization error control. Firstly, to improve the accuracy of geomagnetic matching, we optimize the observations of multi-mode geomagnetic matching (MMGM) by increasing the coverage of magnetic fingerprints. Secondly, we construct a parameter-free stochastic measurement model for the EKF framework by employing variance component estimation, PDR theoretical error, and relative displacements. Thirdly, we propose a multi-level 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.