Non-intrusive People Counting and Identification Simultaneously with Commodity WiFi Devices

Recently, WiFi-based sensing is gaining immense attention in safety monitoring domain for people behavior detection and recognition. The underlying principle of WiFi sensing is that WiFi signal can capture signal changes caused in surroundings and extract the unique signal patterns corresponding to specific behaviors. The capacity allows for the detection,recognition and estimation of behavior attributions. In this paper, we leverage Channel State Information(CSI) to detect individuals entering and exiting doors for counting and analyzing gait behaviors to identify individuals. First, we proposed a sensing-indicator parameter to detect people’s presence and leverage the difference of two antennas to determine whether an individual is entering or exiting. Additionally, we utilize Doppler Frequency Shift(DFS) to estimate the presence of abreast people. Subsequently, we calibrate DFS to estimate stride frequency for gait velocity and stride length. To demonstrate the efficacy of our proposed method, we have designed several experimental schemes. Experiment results show that the detection accuracy of moving people reaches more than 95% in strong sensing zone and 65%-70% accuracy in weak sensing zone. The average accuracy of people counting is 90% and people identification can obtains good performance with less than six volunteers.