Drift-Resistant Heading Estimation for Smartphone-Based Indoor Positioning via Adaptive Calibration Using Wi-Fi Fingerprinting and Magnetic Stability

Accurate heading estimation is essential for improving position accuracy and enabling intuitive navigation by indicating the correct direction to the destination, yet it remains challenging on smartphones because of short term compass bias, long term gyroscope drift, and unconstrained user motion. Previously proposed external infrastructure-dependent methods demonstrate improved heading estimation; however, such reliance faces scalability issues. Conversely, pervasively available Wi-Fi fingerprinting approaches primarily focus on position estimates, overlooking opportunities for heading calibration through temporal continuity and motion-state alignment. This paper utilizes this alignment with the fingerprinting sequence to infer heading calibration. The approach evaluates the suitability of fingerprint positions for calibrating heading using spatial variance and signal strength, incorporating an adaptive thresholding mechanism driven by continuous self-learning to dynamically refine the qualification criteria. A multi-layer validation scheme further improves reliability by cross-correlating displacement-based step estimates and compass-derived headings with corresponding estimates from qualified fingerprinting solutions. To complement this, we propose exploiting Quasi-Static Magnetic Field Intervals (QSMFIs) for reliable heading calibration. Unlike current QSMFI techniques designed for foot-mounted systems with fixed detection parameters, our method introduces a smartphone-oriented method tailored for unrestricted handheld use. A self-attention mechanism is employed to reject faulty QSMFIs, dynamically adjusting window sizes and detection thresholds based on historical magnetic stability, allowing detection across diverse environments and motion contexts. These calibrated headings are integrated within a loosely coupled Extended Kalman Filter that fuses dead reckoning and Wi-Fi fingerprinting for robust position estimation. Real-world experimental results demonstrate that the integrated approach significantly outperforms in reducing drift and enhancing positioning robustness.