In recent years, research on human behavior recognition using Wi-Fi has been active. The advantages of using Wi-Fi include fewer privacy concerns than using video, no need for maintenance such as recharging, and lower installation costs because existing Wi-Fi facilities can be used. However, Wi-Fi is susceptible to environmental influences, and in many cases, learning must be done on an environment-by-environment basis. In addition, most current Wi-Fi-based action recognition research is limited to gesture recognition.
In this research, we aim at activity recognition using environment-independent features. One environment-independent feature is directional radio wave strength. Wision calculates the directional radio wave strength using the phase difference that is geometrically determined based on the direction of arrival and the distance between receiving antennas. Thus, environment-independent imaging is possible. It can be inferred that a person reflecting the radio wave exists in the direction where the value of the radio wave strength obtained by imaging is large. The way in which radio waves are reflected changes as a person acts. Since this reflection pattern is different for each action, the time series of radio wave strength by direction obtained by imaging is also expected to be different for each action. Therefore, we believe that action recognition can be achieved by classifying the time series of radio wave strength by direction using machine learning.
