Passive UHF-RFID Hyperbolic Positioning of Moving Tags by Exploiting Neural Networks

In this paper we propose a novel real-time tracking method of a moving UHF-RFID tag. The purpose is to track the interests of RFID-tagged visitors inside a museum from a set of fixed antenna-installations. Two antenna pairs collect phase-measurements from the target tag. Phase differences are calculated for each pair and then mapped to distance-differences of the target-tag from the two antennas. The latter corresponds to a hyperbola for each pair of antennas. The intersection of the two hyperbolas denotes the position of the tag. The cross section of the hyperbolas is derived by a trained neural network. The proposed method neither requires knowledge of the tag’s initial position nor the trace followed (e.g., conveyor belt). Its computational complexity allows for real-time applicability. Experiments were conducted inside multipath-rich laboratory environments. Two types of experiments were conducted to validate the performance of the algorithm. Firstly a tag was placed on a moving robot, which estimated its own position at cm accuracy, thanks to its lidar sensor, representing the ground truth. Secondly a tag was placed on an ArUco Marker which was carried by a human following various trajectories. The proposed method achieved tracking with mean error under 0.5m throughout the experimental campaigns.