Pedestrian Heading Correction Method Based on HMM Combined with PF

In smartphone-based pedestrian dead reckoning (PDR), heading accuracy is crucial for position estimation, and gyroscope drift and oscillation errors are among the primary limiting factors. To mitigate these effects, Particle Filtering (PF), known for its excellent performance in handling nonlinear problems, is applied to correct quaternion parameter errors derived from gyroscope data, resulting in improved heading estimation. For cases where sudden changes in pedestrian heading induce significant gyroscope vibration noise, the Hidden Markov Model (HMM) is introduced into PF to construct an error state transition matrix (resulting in the PF-HMM filtering method), leveraging HMM’s ability to model errors in a way that closely aligns with physical noise. This approach effectively reduces the impact of gyroscope drift and noise, yielding reliable quaternion estimations and more accurate pedestrian heading direction. Experimental results demonstrate that across various paths, this method effectively suppresses the impact of gyroscope drift on heading accuracy while maintaining high real-time performance. For instance, in a long-distance rectangular path, the proportion of heading deviations greater than 7° using PF alone is 13.20%, whereas PF-HMM, benefiting from the error transition matrix estimated by HMM that better reflects the actual physical noise of the gyroscope, successfully compensates for gyroscope errors, reducing the proportion of deviations greater than 7° to just 2.77%.