WiFi-Based Multiuser Identity, Location, and Activity Recognition Using InceptionTime-Attention Networks

Abstract

Recently, human activity recognition based on WiFi channel state information (CSI) has gained wide attention in fields such as smart homes and e-healthcare. By leveraging the ubiquitous presence of WiFi signals, this technology can identify human activities without the need for additional sensors or cameras, thereby reducing privacy concerns and deployment costs. However, existing studies primarily focus on single-user activity recognition, which is inadequate for real-world scenarios. To address this, this article proposes a novel WiFi CSI model suitable for multiuser activity recognition, identification, and location tracking. We employ preprocessing techniques such as wavelet denoising and sliding window processing to enhance performance on raw data. The newly proposed InceptionTime-Attention deep learning model combines inception modules with attention mechanisms to effectively capture short-term and long-term patterns in WiFi signals. Experiments conducted on the new public dataset WiMANS demonstrate the model’s effectiveness and generalization capability across different scenarios.