Deep Learning Based Pedestrian Dead Reckoning With Vision Anchor Augmentation for Seamless Centimeter-Level Indoor Positioning

In this article, we design a deep learning based pedestrian dead reckoning (DL-PDR) framework with vision anchor augmentation for seamless centimeter-level indoor positioning. The DL-PDR framework integrates deep learning with PDR to address the inherent inaccuracies associated with estimating step lengths and heading directions in PDR methods, as errors in these estimations propagate and accumulate over time, resulting in progressively larger positioning errors. In addition, DL-PDR explores vision-based positioning using existing surveillance cameras while incorporating dynamic anchor augmentation to achieve high-precision real-time localization and correction. According to our review of relevant research, this is the first framework that explores and integrates vision anchor augmentation with inertial data recognition in a deep learning manner to accurately recognize individual walking distances, heading angles, and up-to-date locations, seamless achieving positioning accuracy at the centimeter level. We have implemented an Android-based system with network cameras to validate the feasibility and superiority of the DL-PDR framework. Experimental results show that our framework outperforms existing methods and can accurately recognize the walking distances, moving angles, and located points of individuals with centimeter-level accuracy.