Deep Learning Based Human to Human Interaction Detection Using Wireless Fidelity

Abstract

AbstractRecognizing activity specifically humans, plays a vital role in security alarm systems, elder, fall detection, gesture recognition, smart home systems, and many other fields. Many researchers are being carried out for device-free activity recognition, gesture recognition, elder fall detection, etc. With the increase in technological advancement, the deployment of Wi-Fi has increased. With the presence of Wi-Fi signals almost everywhere in the environment, it helps to detect human activities by the virtue of its property. The Channel State Information (CSI) of the Wi-Fi signal collected contains both amplitude and phase information of Wi-Fi signals. The channel state estimation provides vital information about the channel properties of a communication link which helps in Human activity recognition, as well as human to human interactions. We leverage the change of amplitude in the Wi-Fi signals when there is a human activity or human-to-human interaction. The paper uses the deep learning method which is used to classify five human-to-human interactions in three different phases. The proposed convolutional neural network can identify 5 classes (approaching, departing, high-five, handshake, and hug) of Human-Human Interactions (HHI) with an accuracy of 95.76%.KeywordsHuman activity recognitionDeep learningConvolutional neural networkWi-Fi