Human Activity Recognition Based on Lightweight Programmable WiFi Chips

Since WiFi channel state information (CSI) has higher granularity and privacy, a large number of researchers want to commercialize it. However, most of the existing CSI-based activity identification systems use WiFi routers and terminals to collect CSI, which leads to unstable CSI data transmission links due to the reality that routers are often connected to multiple devices. In addition, many systems use recognition algorithms that are too complex, making it impossible to monitor activities in time due to the delay in system recognition. Therefore, in this paper, we propose a real-time processing and monitoring system based on lightweight programmable chips to collect CSI data and identify human activities. Firstly, we verify the reliability of collecting CSI data using lightweight programmable chips through comparative experiments. Secondly, we propose an activity monitoring model based on automatic threshold adjustment, which is lightweight to achieve real-time monitoring of intruders. In addition, using EMD technology, the system accomplishes activity recognition based on activity frequency with 96% recognition accuracy and good robustness to environmental changes.