A GNN-based indoor localization method using mobile RFID platform

Abstract

The indoor localization system is widely used in many scenarios. Here, radio frequency identification (RFID) technology has been approved to be an effective solution. Many RFID-based localization systems either require high accuracy systems to estimate trajectory such as Synthetic Aperture Radar (SAR)-based methods or are highly depend on the density of reference tags. The current study demonstrates the first application of Graph Neural Network (GNN) for 2D indoor localization using RFID technology. A graph regression approach was used to predict tag coordinates where comparisons were made between three popular GNN models, demonstrating the feasibility of GNNs. The best model achieved a mean absolute error of 5.74cm.