EasyCount: Crowd Counting Based on Easy Deployment Using Commodity Wi-Fi

Recent years have witnessed the great potential of Wi-Fi-based crowd counting technique due to its popularity and non-invasive. However, the existing schemes with high accuracy require labeled data with different numbers of people for a priori learning, which poses an obstacle to the usability of crowd counting systems. Hence in this paper, we propose EasyCount, an easy learning-free multi-environment crowd counting system that only needs one person's prior data for simple calibration. First, we model the reflection signal of an individual as a Gaussian stochastic process. Then, we propose the ratio of dynamic power to noise power (RDN), which establishes a mapping between the autocorrelation function (ACF) of Channel State Information (CSI) amplitude and the number of people. The proposed RDN metric is proportional to the number of people, thus eliminating the dependence on labeled data with different numbers of people. Finally, we design a calibration scheme to obtain the base RDN, which enables our system to overcome the impact of different environments. Extensive experiments show that EasyCount can be easily deployed in different environments, achieving an average accuracy of 93.9 % for 0–6 people counting tasks in multiple environments with only one pair of transceivers, which is comparable to the existing state-of-the-art works.