A Transformer-Based Methodology for Person-Independent Human Activity Recognition Using Wi-Fi Csi

Abstract

The use of Channel State Information (CSI) for human activity recognition holds great promise in healthcare applications, particularly for remote patient monitoring. By interpreting variations in Wi-Fi signals, CSI can be leveraged to detect physical activities, falls, and daily movements. This capability enables the monitoring of patients without relying on wearable sensors or intrusive cameras, offering a fully noninvasive solution. Motivated by this potential, this paper proposes a wireless sensing model called MDA-CSI for recognizing human activities in indoor environments. MDA-CSI employs a Transformer-based architecture designed to process time-series information and effectively capture temporal dependencies. The proposed model is generalizable, allowing it to identify activities performed by individuals who were not included in the training phase.