People Counting with Carry-on RFID Tags.

People counting tracks the number of people entering or exiting any given space in real-time, which helps make intelligent business decisions and improve quality of service (QoS) for a range of commercial applications, e.g., retail analytics, queue management, and occupancy monitoring. Existing CV-based and CSI-based solutions suffer from problems ranging from blind zones and privacy breaches to low robustness and complicated system calibration. In this paper, we propose a new solution to the counting problem with carry-on RFID tags. The basic idea behind the curtain is that when a person is under an RFID reader's coverage zone, his/her carry-on tags can be read by this reader at the same time. We can classify the tags based on the records of reading profiles. Hence, this problem can be boiled down to a clustering problem. To achieve this goal, we build a feature vector for each tag by gathering its reading profile, specify the distance between two feature vectors by combining the Hausdorff distance and the Euclidean distance, and perform a maximum density clustering algorithm to get the distributions of each tag. Afterward, we design an outlier detection algorithm to screen out the cluster centers from the non-cluster ones. The number of cluster centers is treated as the number of people. We implement the RFID-based counting model in a commodity RFID system. Extensive experimental results show that our approach is superior to existing solutions under different scenarios.