RVTrack: Identity Tracking by RFID-Vision Fusion with Small Training Cost

Person identification and tracking (PIT) from cameras is an essential application of computer vision (CV). Traditional PIT solutions require a labor-intensive training process involving photographing and manual annotation. This paper presents a PIT method that eliminates the need for such training efforts by leveraging radio-frequency identification (RFID) technology, requiring only that each person carry a low-cost RFID tag. The proposed system, RVTrack, is an RFID and CV fusion system that labels individuals in camera footage with the ID from their RFID tag. RVTrack includes a scheme to effectively differentiate persons with similar appearances, which existing RFID-CV fusion solutions are difficult to achieve. Experimental results demonstrate that RVTrack can actively identify and track the RFID-tagged target objects with an accuracy of 90.14% using commercial RFID devices and cameras, even in complex indoor environments where various multipath reflectors exist.