Office and Home Environment Security Applications of Wi-Fi Sensing Via Deep CSI Processing

Device-Free Wireless Sensing is an emerging field of interest and has gained traction in health monitoring and security measures. The presented approach aims to overcome privacy issues, blind spots in the non-line-of-sight, and poor light conditions while processing camera images using computer vision techniques. In this research, a novel method has been proposed in order to extract features from Channel State Information (CSI) samples gathered via visual classification. As a base model, VGG16, a convolutional neural network model, for large-scale image recognition, which has been used in many deep-learning image classification problems in different fields of study, has been used. Although it is slow to train, it is easier to implement. The presented method is based on an offline process that consists of the following steps: 1) collecting CSI samples, 2) generating four graphs (amplitude, channel frequency response, phase, and channel impulse responses of all 30 subcarriers), 3) replaying them through time and capturing images every second, label images based on activity, 4) passing the images to a model and training it to classify activities, and finally 5) testing the accuracy against a test dataset. VGG16 has been used as a base model, achieving an accuracy of 97% in an indoor environment with up to 5 people.