Device-Free Sensing for Gesture Recognition by Wi-Fi Communication Signal Based on Auto-encoder/decoder Neural Network

Gesture recognition has been found to be a vital mission for a variety of applications, such as smart surveillance, elder care, virtual reality, advanced user interface, etc. Recently, an emerging sensing technology, namely device-free sensing (DFS), has been introduced to the domain of gesture recognition which only uses radio-frequency (RF) signals without the need to equip any devices or extra hardware support; thus, it would be a natural choice to fully leverage ubiquitous Wi-Fi signals in almost every modern building. Although the feasibility of using this technology for gesture recognition has been explored to some extent, we observe that it still cannot perform promisingly for some gestures which maybe look nearly identical in a certain instant. Therefore, in this paper, we conduct experiments with several typical hand gestures in the opposite direction based on a proposed Auto-Encoder/Decoder (Auto-ED) deep neural network to address gesture recognition in our case. Compared with several traditional learning methods, experimental results demonstrate that our proposed approach can best tackle the challenge of gesture recognition for identical motions, which indicates its potential application values in the near future.