Channel State Information-based Human Activity Recognition System for Internet of Things Using Convolutional Neural Network

This paper presents a channel state information (CSI)-based human activity recognition (HAR) system for the Internet of Things (IoT), which aims to improve the feasibility of the HAR system in the real environment. To this end, the proposed system conducts the following three steps: 1) data acquisition, 2) data preprocessing, and 3) data analysis. In the first step, a capture device extracts CSI from the subcarriers of the wireless signals transmitted by a signal generator. In the second step, the analysis server removes the noise in the CSI dataset and calculates the amplitude for each CSI in the CSI dataset. Finally, in the third step, the convolutional neural network (CNN) model classifies human activities by analyzing the preprocessed CSI dataset. The implementation results showed that our system obtains accurate HAR results in the real environment.