Wifi-based location-independent human activity recognition and localization using deep learning

Abuhoureyah, Fahd Saad Amed (2024) Wifi-based location-independent human activity recognition and localization using deep learning. Doctoral thesis, Universiti Teknikal Malaysia Melaka.

	Text (24 Pages) Wifi-based location-independent human activity recognition and localization using deep learning (24 pages).pdf - Submitted Version Download (578kB)
	Text (Full Text) Wifi-based location-independent human activity recognition and localization using deep learning .pdf - Submitted Version Restricted to Registered users only Download (26MB)

Abstract

Detecting human activities holds paramount significance across diverse domains, encompassing healthcare, security, autonomous driving, and human-computer interaction. Leveraging wireless signals for activity sensing exploits the intricate influence of human activities on signal propagation phenomena such as reflection, diffraction, and scattering. Wireless signal-based human sensing, mainly through WiFi and radar technologies, presents notable advantages, including device-free sensing, resilience to environmental factors, obstacle penetration, and preservation of visual privacy. This work addresses the inherent challenges in WiFi-based Human Activity Recognition (HAR) by focusing specifically on critical aspects: the location dependency of WiFi sensing and the impact of multi-user interactions on signal reliability. Existing HAR systems encounter difficulties recognizing human activities due to variations in physical environments and the complexities introduced by multiple users within WiFi signals. The study aims to advance methodologies to mitigate challenges arising from environmental dependency and multi-user effects, enhancing the precision and adaptability of WiFi-based HAR systems for reliable and robust performance across diverse environmental contexts. The research highlights the role of Deep Learning methodologies in addressing challenges and advancing the capabilities of HAR technology. First, we employ the advanced Seq2Seq Recurrent Neural Network (RNN) technique to achieve high accuracy in HAR with few layers of the Long Short-Term Memory (LSTM) algorithm. Precise activity recognition and incorporation of through-wall sensing capabilities are achieved within the deep learning framework. Second, Multi-head Attention Mechanism Networks capture intricate patterns in Channel State Information (CSI) data, enhancing recognition accuracy for human activities detected through WiFi signals. Third, recognizing the capability of location independence, we propose a novel locationindependent HAR using a self-learning CSI-based technique for wireless sensor networks. This innovative approach reduces the impact of environmental factors on HAR accuracy, ensuring robust performance across diverse spatial contexts. Fourth, addressing the challenge of human interaction recognition in multi-user environments, WiFi signal processing with Independent Component Analysis (ICA) and Continuous Wavelet Transform (CWT) techniques is introduced. An efficient real-time localization method is introduced in the fifth section, which achieves location-independent localization by utilizing the fusion of the Received Signal Strength Indicator (RSSI) and CSI. The fusion contributes to the development of reliable and adaptable HAR systems across varying environmental contexts. A trajectory mapping approach using CSI-Triangulation with deep learning is proposed to refine the localization capabilities of WiFi-based HAR, offering an accurate and robust solution for localization in diverse real-world scenarios. The adaptive strategy accommodates variations in signal characteristics and environmental factors, highlighting the robustness of the presented methods in scenarios involving various user interactions and environmental conditions. The findings contribute to the improvement of HAR and localization systems and have acheived high accuracy of clasification up to 97.5% with enhancement of 6% of localization and tracking accuracy.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Location-based services, Signal processing
Divisions:	Faculty Of Electronics And Computer Technology And Engineering
Depositing User:	Norhairol Khalid
Date Deposited:	24 Jun 2025 02:19
Last Modified:	24 Jun 2025 02:19
URI:	http://eprints.utem.edu.my/id/eprint/28837
Statistic Details:	View Download Statistic

Actions (login required)

View Item