RISAR: RIS-assisted Human Activity Recognition with Commercial Wi-Fi Devices

Human activity recognition (HAR) is integral to smart homes, security, and healthcare. Existing systems face challenges such as the absence of line-of-sight links, insufficient details regarding the sensing subject, and inefficiencies in noise reduction and feature extraction from channel state information. To address these issues, this study builds a reconfigurable intelligent surface (RIS)-assisted passive human activity recognition (RISAR) system, compatible with commercial Wi-Fi devices. RISAR employs a RIS to augment the orientation and spatial diversity of Wi-Fi signals, thereby facilitating improved detection capabilities. A new denoising and feature extraction technique, the high-dimensional factor model, based on random matrix theory, is proposed for noise elimination and salient temporal feature extraction. On this basis, a dual-stream spatial-temporal attention network model is developed for assigning variable weights to different characteristics and sequences, mimicking human cognitive processes in prioritizing essential information. Experimental analysis shows that RISAR significantly outperforms existing HAR systems in accuracy and efficiency, achieving an average accuracy of 97.26 system but also underscore its potential as a robust solution for activity recognition across complex environments.