Poster: Activity Recognition Using CSI Backscatter with Commodity Wi-Fi

Recently, there is growing interest in Wi-Fi CSI-based activity recognition due to its low setup costs. However, accurate CSI-based activity recognition depends on the number of Wi-Fi devices, which is suboptimal cost-wise. Our proposed solution is to use low-power backscatter tags within a Wi-Fi CSI sensing system, collecting multiple CSI data streams from various Wi-Fi channels. This enhances the number of observations without the need to install a large number of Wi-Fi devices. We evaluated classification of five daily activities using traditional Wi-Fi CSI and backscattered CSI, finding an accuracy improvement by combining them.