Dark Roads, Cloudy Skies, Textureless Walls: ReLoc-PDR Still Finds the Way

For smartphone-based navigation in visually impaired contexts, this study presents ReLoc-PDR, a strong pedestrian positioning framework. In an incremental smoothing graph optimization approach, the system combines pedestrian dead reckoning (PDR) with deep feature-based visual relocalization. The Tukey robust kernel is used to effectively reduce trajectory drift and eliminate aberrant localization errors. Tests in three challenging scenarios—textureless hallways, overcast outdoor situations, and nighttime settings—show that ReLoc-PDR consistently works better than inertial-only PDR, VINS-Mono, and dynamic-weighted PDR/vision techniques. With decimeter-level accuracy and smooth, drift-free trajectories, the suggested architecture allows for dependable pedestrian navigation both indoors and outdoors without the need for GPS.