Device-free Human Activity Recognition Based on CSI Characteristic

In recent years, device-free wireless location and action recognition have increasingly become emerging technology. It does not require carrying any equipment while protects personal privacy, which can enable intelligent perception of human body through shielding effect of the target in wireless network. Since human body has a significant reflection effect on wireless signals, it is possible to estimate a person's position and activities by monitoring changes in the signals. Since wireless signals do not contain annotation information, supervised learning cannot be directly performed on the collected channel state information (CSI). Therefore, we propose a device-free human activity recognition scheme based on CSI characteristic, where we create category labels through One-Hot encoding. Specifically, the CSI amplitude matrix and phase matrix of wireless signals are directly processed to reduce calculation errors and improve classification accuracy. Finally, through experimental comparison, the effectiveness of proposed method in terms of recognition accuracy is demonstrated.