People Counting and Gesture Recognition System Using Indoor Wi-Fi Sensing

This paper proposes a human activity recognition (HAR) system that detects the number of people and the corresponding gestures they are performing through Wi-Fi channel state information (CSI). Existing CSI-based gesture recognition (GR) systems are only limited to one person, while CSI-based crowd-counting (CC) systems do not detect gestures or human activity. The CSI data is collected using an off-the-shelf Raspberry Pi single-board computer (SBC) with Nexmon CSI firmware over four days with several person-gesture combinations and trials. Support vector machine (SVM) models with different kernels were trained using CSI data and performed well, particularly in the two-person gesture recognition scenario, with up to 93% accuracy.