Factor Graph-Based Geomagnetic SLAM Approach for Smartphone-Based Indoor Localization
TL;DR
A factor graph-based geomagnetic SLAM method combining confidence-aware loop closure detection with local–global collaborative optimization combining confidence-aware loop closure detection with local–global collaborative optimization to support simultaneous mapping and localization.
Abstract
Indoor geomagnetic positioning provides a low-cost solution for smartphone-based localization without requiring additional infrastructure. However, conventional geomagnetic positioning methods depend on labor-intensive offline magnetic data surveys, limiting deployment flexibility. To address this problem, this paper proposes a factor graph-based geomagnetic SLAM method combining confidence-aware loop closure detection with local–global collaborative optimization. A Gaussian process regression (GPR)-based magnetic model is constructed from measured step-level magnetic observations, and a window-based particle filter integrates magnetic sequence matching with pedestrian dead reckoning (PDR). Confidence-aware loop detection combines magnetic sequence similarity with PDR-derived geometric verification, while continuous confidence weights regulate the influence of accepted associations on back-end optimization. Factor graph-based local–global collaborative optimization coordinates trajectory correction with GPR map refinement to support simultaneous mapping and localization. Cross-user experiments involving three participants yielded mean localization errors of 1.19–1.31 m and RMSEs of 1.35–1.45 m. These results demonstrate that the proposed framework reduces accumulated trajectory drift while supporting online geomagnetic mapping.