Radio-frequency (RF) sensing for deep awareness of human physical status

Abstract

In the last few years, there has been a noticeable trend of utilising Radio Frequency (RF) such as Wi-Fi signals coupled with deep learning techniques for Human Activity Recognition (HAR), which is less invasive and contactless. Changes in signal transmission can be detected by analysing the Channel State Information (CSI) extracted from received Wi-Fi signals. This study explores gesture recognition using CSI data processed through deep learning models. Three models—BiLSTM, CNN-BiLSTM, and Transformer—were evaluated on key performance metrics, including accuracy, precision, recall, and F1-score. The Transformer model demonstrated superior performance across all metrics, achieving an accuracy of 92.24%, precision of 92.56%, recall of 92.23%, and F1-score of 92.17%. These results highlight the effectiveness of Transformer architecture in capturing complex patterns in CSI data, outperforming the other BiLSTM and CNN-BiLSTM approaches. This work hopes to provide insights into the use of deep learning techniques and architecture for robust gesture recognition applications.