Human Activity Recognition With Commercial WiFi Signals

Abstract

The next generation of mobile communication aims to extend the capabilities of traditional communication by reshaping the environment with wireless signals. The channel state information can describe the propagation characteristics in wireless communications, which is beneficial in developing wireless communication networks towards intelligent communication and wireless sensing networks. Raspberry PI with Nexmon firmware patched can extract the channel state information from WiFi signals and realize human activity recognition. However, the phase values on some carriers are susceptible to noise, resulting in phase errors after singular value decomposition. To solve this problem, a method is proposed in this paper to find the optimal phase value by dynamic time warping algorithm utilizing the property of orthogonality between amplitude and phase. In contrast to the conventional recognition strategies, the proposed optimal phase extraction method with commercial WiFi signals can further improve the accuracy of the recognition strategy under different complicated scenarios.