Utilizing RFID Tag Motion Detection in High Tag Density Environments for Customer Browsing Insights

• Published in: IEEE journal of radio frequency identification (Online) Vol. 5; no. 4; pp. 345 - 356
• Main Authors: Wan, Chieh-Yih, Tanriover, Cagri, Shah, Rahul C.
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.12.2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Customers; Density; Motion detection; Motion perception; Phase estimation; Radio frequency identification; Radiofrequency identification; Received signal strength indicator; Retail stores; RFID tags; Security management; Tags
• ISSN: 2469-7281; 2469-7281; 2469-729X
• DOI: 10.1109/JRFID.2021.3087229

"Responsive retail" is becoming a key differentiator for brick-and-mortar retail stores to remain competitive with online alternatives. Growing RFID deployments in retail positions RFID as a compelling means to capture customer-item interactions, giving retailers insights to optimize store layout, item arrangements and even strategic item placements to increase sales. Such insights depend on detecting motion of a few tags among thousands of stationary tags, which is an open challenge. We have created a fully FCC compliant UHF RFID testbed with 1000 tagged clothing items. By collecting customer-item interaction data in realistic retail environments, we show existing algorithms working in low tag density environments (i.e., < 100 tags) start to fail in denser environments with a few hundred tags or more. To address this gap, we developed a motion detection algorithm using RSSI and phase information from tag reads to accurately discriminate a few moving tags among hundreds of stationary tags (up to 1000) in near real-time. Our algorithm runs on standard RFID systems without modifications, and with acceptable processing overhead. Our results cover two different RF environments, presence of multiple people, simultaneous interaction with multiple items and different tag densities and achieve >90% accuracy in representative retail environments.