Loading...

Human Activity Recognition Based on Wi-Fi Channel State Information using Deep Learning

Samavati, Ali | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57750 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Karbasi, Mohammad
  7. Abstract:
  8. With the advancement of wireless technologies and sensing methods, numerous studies have demonstrated the successful reuse of signals from networks based on the IEEE 802.11 standard for human activity recognition. Recently, Channel State Information, available in newer commercial modems, has been widely used in research for achieving more accurate results. However, most studies collect and analyze samples by generating artificial traffic, such as downloading large files, short-interval pinging, or altering settings to increase beacon signal transmission rate. This approach contradicts the normal operating conditions of these networks. Additionally, in such studies, the transmitter and receiver are typically positioned in fixed and separate locations. In contrast, in real-world home networks, a fixed access point usually exists, while mobile devices such as phones, laptops, or tablets connect to it dynamically. In this thesis, to address the first issue, only beacon signals from a single access point with standard settings and low transmission rates are utilized. These signals are consistently transmitted at regular intervals by access points. To address the second issue, a real access point is implemented for the first time using software-defined radio, enabling it to receive its own transmitted signals, akin to an active radar. Given the capability of receiving its own transmitted signals, this access point can independently analyze the information without affecting the network or requiring additional equipment. Channel State Information was collected from the access point’s own beacon signals for 9 volunteers performing 6 types of activities. This research achieved an accuracy of approximately 80% in activity recognition using a hybrid neural network consisting of Convolutional Neural Network and Long Short-Term Memory models
  9. Keywords:
  10. Human Activity Recognition ; Channel State Information (CSI) ; Long Short Term Memory (LSTM) ; Convolutional Neural Network ; Wireless Sensor Network ; Wi-Fi Access Point ; Beacon Signal

 Digital Object List

 Bookmark

No TOC