Human Segmentation Using Commercial Wi-Fi Device

Image segmentation technology is widely used in automatic driving, smart home and other fields. However, due to the problem of privacy of image methods, they face restrictions when used in private occasions. In order to avoiding private problem image-based method, we proposed a human segmentation method based on commercial Wi-Fi devices named Wi-Seg, which is used to extract human silhouette information and generate pixel-level annotations. So that it can be integrated with existing human- computer interaction schemes using image segmentation, and avoiding the privacy problem of image segmentation methods. In this paper, the channel state index (CSI) in the Wi-Fi signal is collected as the original data, and a encoder-decoder deep neural network is designed to extract the human silhouette information from the CSI signal. Compared with the mainstream image segmentation methods, our Wi-Seg is superior to recent work that using Wi-Fi signal to extract human silhouette information, and performs as well as image segmentation methods.