Discriminating stray RFID tag reads

In RFID systems, it is unavoidable that we read stray tags occasionally and it is critical that we can reliably filter out stray tag readings for accurate tracking. In this paper, we studied a set of features namely RSSI Euclidean distance (RED), number of reads Euclidean distance (NED) and time between reads Euclidean distance (TED) for classifying RFID reading into target and stray class membership. Comparison of these features for their abilities to discriminate stray tags from target tags showed that RED and TED are stronger than NED. Performance measures based on the k-Nearest Neighbour classifier shows that RED and TED dual feature space outperforms in terms of correctly classifying true target and true stray tags.