Proximity Detection and Trajectory Recognition with Machine Learning for UHF RFID Systems

We report on the proximity detection of a moving object tagged with a passive Ultra High Frequency (UHF) Radio Frequency Identification (RFID) tag compared to a stationary tag and the recognition of predefined movement trajectories of tagged objects. Both methods are based on sequentially received tag responses with time and reader information. The recording system consists of a commercially available UHF RFID reader and a switched antenna system with eight antennas. We use supervised machine learning classification methods comprising convolutional layers, Decision Tree (DT), Random Forest (RF), and Ridge classifiers. The proximity detection is compared with two decision thresholds and shows accuracies of more than 83 percent. Evaluation of the trajectory recognition shows an accuracy of 87.5 percent.