Machine Learning-based Inversion of Wireless Signals for Real-Time Gesture Recognition.

This work presents the latest results on the development and experimental validation of a human-machine interaction systems based on the use of Wi-Fi commodity devices and the processing of the channel state information (CSI) for gesture and hand-pose recognition. The inversion method is based on a customized machine learning strategy that allows guaranteeing the real-time operation as well as high percentages of estimation accuracy and robustness. The reported experimental validation, carried out within a real house environment, verifies that the system can recognize four different arm/hand poses with accuracy higher than 95%.