A probabilistic approach for cleaning RFID data

The inherent uncertainty in RFID signals requires an RFID middleware system to clean the input data after capturing. Typically these systems employ a low pass filter for reducing errors. In this paper we propose an approach for data cleaning that exploits basic characteristics of RF signals as well as maximum likelihood operations. With our filter we improve proximity detection of RFID tags. Our solution enables reasoning about the position of RFID tags in the reader´s range without measuring the signal strength of tag responses. It is therefore applicable on top of standard reader interfaces. Our solution improves data cleaning wherever the tag to reader distance is relevant. For instance this enables correct ordering of items that pass by a reader on a conveyor or enhances tracking scenarios with RFID equipped fork lifts. We demonstrate the benefits of our approach compared to low pass filtering.