Wi-Fi-Based Human Activity Recognition Using Convolutional Neural Network

EasyChair Preprint 6273

Abstract

Unobtrusive human activity recognition plays an integral role in a lot of applications, such as active assisted living and health care for elderly and physically impaired people. Although existing Wi-Fi-based human activity recognition methods report good results, their performance is susceptible to changes in the environment. In this work, we present an approach to extract environment independent fingerprints of different human activities from the channel state information. First, we capture the channel state information by using the standard Wi-Fi network interface card. The channel state information is processed to reduce the noise and the impact of the phase offset. In addition, we apply the principal component analysis to removed redundant and correlated information. This step not only reduces the dimensions of the data but also removes the impact of the environment. Thereafter, we compute the spectrogram from the processed data which shows the environment independent fingerprint of the performed activity. We use these spectrogram images to train a convolutional neural network. Our approach is evaluated by using a human activity data set collected from 9 individuals while performing 4 activities (walking, falling, sitting, and picking up an object). The results show that our approach achieves an overall accuracy of 97.78%.

Keyphrases: Channel State Information, Convolutional Neural Network, Principal Component Analysis, Spectrogram

@booklet{EasyChair:6273,
  author    = {Muhammad Muaaz and Ali Chelli and Matthias Pätzold},
  title     = {Wi-Fi-Based Human Activity Recognition Using Convolutional Neural Network},
  howpublished = {EasyChair Preprint 6273},
  year      = {EasyChair, 2021}}