Towards People Counting Using Wi-Fi CSI of Mobile Devices

Monitoring crowdedness in public spaces such as shopping malls is useful for various services like marketing, safety, evacuation planning. For infrastructure-based people counting, Wi-Fi Channel State Information (CSI) has attracted attention as it does not require any additional deployment. However, the existing approaches assume multiple fixed infrastructures such as Wi-Fi access points (APs), which is different from standard AP deployment since they are installed for wireless network services. To solve this problem, our goal is to estimate the number of people (i.e. people counting) by using mobile devices such as smartphones and a small number of fixed APs. Since CSI represents the difference between the amplitude and the phase of the transmitted and received radio waves, we can estimate the change of the propagation environment from CSI. However, CSI is very noisy due to various factors that are not related to the number of people. In this paper, we focus on Sampling Frequency Offset (SFO), Carrier Frequency Offset (CFO), A/D Converter (ADC) delay, and quantization error, and design methods to mitigate their effects. Then, we propose people counting by using CSI variance as a location-independent feature for machine learning. From the evaluation results, we have confirmed that our method achieves the Root-Mean-Squared-Error (RMSE) of 0.49, which is much better than RMSE of 3.7 without denoising,