Device-Free Cross-Environment Human Action Recognition Using Wi-Fi Signals

The research of human action recognition (HAR) based on Wi-Fi signals shows great application value in fields of human-computer interaction. However, many existing Wi-Fi-based HAR systems are vulnerable to environment-variant and show poor generalization capabilities in new environments. To solve this problem, in this paper, a cross-environment HAR system (Wi-CHARS) based on channel state information (CSI) of Wi-Fi signals is proposed. At first, according to the characteristics that human activities have different influences on different subcarriers of CSI, a dynamic data detection method called (DDDM) is proposed for the data segmentation. After that, a HAR adversarial network is designed to realize the cross-environment HAR, with the adversarial learning strategy, the network can learn to extract environment-independent action features by reducing the action feature distribution distance of different environments, thus realizing good cross-environment HAR performance. The results of experiments show that the proposed system achieves more than 80% HAR accuracy in new environments.