Harnessing Wi-Fi CSI for Enhanced Human Activity Recognition: A Feature-Driven Approach

Wi-Fi routers are ubiquitous in modern households, offering the capability to estimate Wi-Fi channel state information (CSI), which can be harnessed to sense the surrounding environment. CSI captures intricate signal variations resulting from human movement, such as amplitude, phase, and Time of Flight (ToF). This paper proposes a Human Activity Recognition (HAR) method based on a dataset of CSI features. The features were extracted across multiple dimensions, including time-domain, frequency-domain, statistical, temporal, and spatial characteristics. Feature selection techniques, such as correlation analysis, were employed to reduce data dimensionality. Machine learning models were trained on the refined feature set for indoor environments in line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios. Our method significantly outperformed the raw Wi-Fi CSI dataset-based approach in HAR classification accuracy. The results demonstrate the effectiveness of various classifiers in HAR using Wi-Fi CSI feature data. This study provides critical insights into the application of feature extraction and selection techniques for precise activity recognition, with potential applications in healthcare, security, and human-computer interaction.