Adaptive People Counting and Localization Based on Meta Learning with WiFi NICs

The paper focuses on utilizing channel state information (CSI) from commodity WiFi network interface cards (NICs) for people counting and localization systems. CSI offers informative data on signal characteristics in amplitude and phase, but it is prone to offsets by hardware imperfection and co-channel interference from other WiFi devices. The paper introduces a preprocessing method to remove offsets, ensuring low-latency operation without additional filtering. Then, we propose people counting and localization models that are based on meta learning to adapt to different environments. Numerical results demon-strate the effectiveness of the proposed meta learning models in achieving high sensing accuracy compared to other simpler training methods.