Personnel Identification System Using the Wireless Channel Characteristics and Machine Learning

Abstract

In the presence of a Wi-Fi network, movements of individuals lead to changes in the channel state information (CSI) of the Wi-Fi network. Thus, it is possible to use CSI to identify specific individuals based on walking gait and under movements. We have built a hardware system to collect CSI and developed CSI real-time acquisition and visualization software. We further improve results by preprocessing the CSI to remove noise by outlier removal, a low-pass filter, and a discrete wavelet transform. For identity recognition, we propose a parallel learning structure using a convolutional neural network (CNN) and bidirectional long-term and short-term memory network to extract amplitude and time sequence features of human gait at the same time. We then classify the two gait features using Softmax function to identify individuals. Our identity recognition experiment with 30 people yielded a maximum accuracy of 98.7% with our device and software. The system has a value of application in the field of smart home and intrusion detection.