Adaptive multi-sensor fusion for pedestrian indoor positioning using UCR-DTW, improved CRF, and SAGE-HUSA filtering

Indoor positioning in smart buildings is challenging because satellite signals are obstructed, while inertial navigation methods suffer from sensor noise, magnetic interference, and cumulative errors. This study proposes a high-precision indoor positioning method integrating Pedestrian Dead Reckoning (PDR), map matching, and multi-sensor fusion for real-time pedestrian positioning. The proposed framework employs acceleration and geomagnetic features for stationary state detection, Dynamic Time Warping (DTW) for adaptive gait recognition, and a robust adaptive Kalman filter for heading estimation. Building topology constraints are incorporated through an improved Conditional Random Field (CRF), which is combined with University of California Riverside-optimized DTW (UCR-DTW) for geomagnetic sequence matching. Finally, a Sage-Husa filter fuses multi-source information to generate robust position estimates. Experiments on the IPIN 2025 Competition Track 3 (Smartphone-based) indoor positioning dataset show that the proposed method achieves a mean positioning error of 0.45 m and a trajectory integrity rate of 98.2%. The improved CRF reaches a map-matching accuracy of 96.8% after six iterations and maintains 71.6% accuracy under level-5 magnetic interference. These results demonstrate that the proposed framework effectively suppresses cumulative errors and magnetic disturbances while constraining trajectories within navigable areas, providing accurate and reliable real-time positioning for pedestrian indoor navigation. Future work will incorporate height information during multi-storey transitions to further improve cross-floor positioning performance.