Transformable Fingerprinting with Deep Metric Learning Approach for Indoor Localization

Within state-of-the-art indoor localization approaches, fingerprinting based method is more applicable and easier to integrate into most of today's commodity Wi-Fi devices such as mobile phones and IOT devices which require low cost and computation burden. However, most fingerprinting systems intrinsically depend on fixed channel propagation environment and thus suffers huge reconstruction cost and high localization error when environment changes. In this paper, we propose a novel transformable fingerprinting localization method based on deep metric learning approaches. Our fingerprinting reconstruction method only requires some fresh measurements of CSI (Channel State Information) on a few reference points (RPs) with all the outdated CSI fingerprinting. Extensive system level simulations on Quadriga show that an average of 0.2m error reduction is achieved when our reconstruction method is applied.