A multi-hypothesis particle filtering approach for pedestrian dead reckoning
Résumé
Map aided Pedestrian Dead Reckoning (PDR) methods seek to guarantee an autonomous positioning solution for pedestrians. Inertial sensors, assembled in one measurement unit, are usually fused with map data in order to palliate the well-known drift errors and step detection limitations, especially indoors as no GNSS signal is available. In addition to these remaining limitations, the measurement unit carrying mode has also to be accounted for in order to correct the walking directions, which is still an open issue in the literature. The aim of this study was to handle the above-mentioned difficulties within the framework of handheld devices; a magnetic and inertial measurement unit was used during experiments. First a routing graph was built according to the pedestrian's possible motions, depending on his mobility profile and surroundings, and second, an adaptive decision process was developed to determine which data among GNSS and the graph elements should be used to assist the location process; a particle filter was implemented to achieve these objectives. The main findings of this study can be summarized in three points. First, the step length model can be calibrated using GNSS within the straight phases of the subject's walk, preferably using particle filters because they offer a greater freedom in the positioning configuration; second, the misalignment between the walking direction and the pointing direction of the measurement unit is assessable in the geometrically constrained areas; and finally, the space configuration can be exploited to make a convenient choice of the external measurements to be considered.