A hidden Markov model for distinguishing between RFID-tagged objects in adjacent areas

Abstract

Distinguishing between RFID-tagged objects within different areas poses an important building block for many RFID-based applications. Existing localization techniques, however, often cannot reliably distinguish between tagged objects that are close to the border of adjacent areas. Against this backdrop, we present a hybrid approach based on an ANN and a HMM that leverages not only low-level RFID data streams but also information about physical constraints and process knowledge and thus incorporates scene dynamics. We experimentally demonstrate the performance of our approach considering a RFID-based smart fitting room which is a practically relevant application with limited process control in an environment with strong multipath reflections and non-line-of-sight effects. Our results show that our approach is able to reliably distinguish between tagged objects within different cabins. This includes objects hanging on coat hooks at partition walls of adjacent cabins, i.e., at a maximum distance of 5 centimeters to the border of an adjacent area.