Activity Identification via Wi-Fi Channel State Information with Neural Networks

Abstract

Leveraging Wi-Fi Channel State Information (CSI) offers a fresh approach to environmental sensing and detecting human activities (HAR). This technique has diverse safety and security applications, utilizing existing Wi-Fi routers instead of privacy-intrusive visual methods. This study presents a comprehensive pipeline for human activity recognition via Wi-Fi CSI, comparing two deep learning techniques. We explore the impact of hardware setups on Wi-Fi CSI signals and enhance data collection by seamlessly integrating real-life scenarios, improving model assessment. We investigate Inception Time and LSTM models for activity recognition. The code and dataset are publicly accessible for transparency and community-driven research advancement.