AC-HMM: azimuth-constrained hidden Markov model-based map matching for robust pedestrian positioning in urban canyons

Location-Based Services (LBS) have become integral to contemporary urban life, with Global Navigation Satellite System (GNSS) modules in smart devices serving as the primary geospatial data source. However, urban canyon scenarios challenge satellite positioning. Signal obstruction and multipath effects caused by high-rise buildings result in substantial positioning errors, potentially misplacing users on the sidewalk across the street and compromising LBS reliability. While Map Matching (MM) techniques have proven effective in enhancing positioning accuracy for vehicular navigation systems, pedestrian map matching encounters two primary challenges. First, unlike vehicular roads with directional constraints, bidirectional pedestrian networks hinder direct direction-based road matching. Second, pedestrian pathways, typically situated closer to surrounding buildings, lead to larger GNSS positioning errors, rendering distance-based matching more error-prone. To address these limitations, this paper introduces an Azimuth-Constrained Hidden Markov Model (AC-HMM) based pedestrian map matching methodology. Initially, a comprehensive pedestrian road network map is constructed using open-source datasets. Subsequently, the proposed AC-HMM algorithm is applied to pedestrian map matching, integrating short-term and long-term azimuth information to enhance MM accuracy in complex urban environments. Experimental validation using real-world pedestrian trajectory data demonstrates that the traditional HMM-MM algorithm decreases mean positioning error by approximately 7.1% compared to raw GNSS results. Notably, the AC-HMM-MM methodology substantially outperforms conventional HMM-based map-matching approaches, achieving a 20.7% improvement in matching accuracy and a 37.3% enhancement in mean positioning precision. The algorithm effectively mitigates opposite sidewalk positioning errors typical of high-density urban canyon environments while maintaining real-time processing capabilities for pedestrian navigation applications. Despite these performance improvements, the method preserves computational efficiency with only a minimal 1.27% increase in processing time.