Context-assisted personalized pedestrian dead reckoning localization with a smartphone

Abstract

Pedestrian Dead Reckoning (PDR) plays an important role in multi-sensor fusion of indoor positioning due to its autonomy and continuity advantages. The robustness of PDR significantly impacts indoor positioning accuracy, but various pedestrians and mobility contexts pose challenges for reliable step detection and accurate step length estimation. This paper proposes a context-assisted personalized PDR localization solution to address these challenges. Firstly, by exploiting temporal and frequency domain features, an enhanced step detection method is developed to mitigate false step detection, especially during unfavorable actions of pedestrians. Subsequently, a personalized step length model is proposed, and its parameters are dynamically updated online using other high-precision sensors available within a multi-sensor fusion positioning solution. Moreover, the personalized step length model is further refined using mobility context knowledge. Finally, a novel context-assisted pedestrian velocity model is established for PDR localization to enhance positioning accuracy, particularly when there are changes in mobility contexts. The results demonstrate that the robustness of step detection is improved, and the false detection rate is reduced from 13% to 2%. For various smartphone users, the proposed context-assisted personalized step length model exhibits a relative error of 2.01%, in contrast to 7.06% observed with the traditional flat model. Consequently, the accuracy of walking distance is enhanced from 92.2% to 98.9%, and the PDR localization error is reduced from 2.49 m to 1.63 m. Importantly, the proposed solution exhibits more robust and consistent performance across different pedestrians, smartphone models, and challenging mobility contexts.

Keywords

• Pedestrian Dead Reckoning (PDR)
• step detection
• mobility context
• indoor positioning
• smartphone