WiFi-based Contactless Gesture Recognition Using Lightweight CNN

Gesture recognition has the potential to become a part of contactless interactions with devices to improve accessibility and ease with applications. As the presence of portable devices remains standard, WiFi will continue to constantly connect these devices. Leveraging this availability, instead of relying on installing special sensors, ubiquitous WiFi sensing devices can decipher motion, thus mitigating additional costs. We develop a low-cost hand gesture recognition system utilizing Channel State Information (CSI) from a few subcarriers in prevalent WiFi signals. This information is sent through a lightweight signal segmentation algorithm and Convolutional Neural Network (CNN) that learns the gestures and successfully distinguishes them. Computationally demanding feature extraction is avoided as it increases processing time and does not scale well with additional gestures. Our model obtains an 96% accuracy rate across three different gestures on average.