Evaluation of Feature Attributes for an RFID Conveyor Belt Application using Probabilistic Models

RFID conveyor belt applications are frequently found in fashion and retail logistics where individually tagged goods are packed into boxes for easy handling. Based on the item-level tagging of goods, these applications are used to identify individual tags and to perform the assignment of tags to their enclosing box. Several aspects such as the extended read range of UHF systems and multipath propagation degrade the detection performance. In this paper we discuss different feature attributes available to the underlying classifier within the RFID system. In particular we address the impact of using the phase information provided by the reader on the classification performance of the RFID system.