| dc.description.abstract | This dissertation delves into the potential of WiFi-based sensing for Human Activity Recognition (HAR), a promising avenue for smart healthcare and remote monitoring due to WiFi's ubiquity and non-intrusiveness. It confronts the technical intricacies of employing Channel State Information (CSI) signals within deep learning frameworks for HAR, addressing critical challenges such as signal variability due to environmental factors, signal noise, interference, and the complexity of mapping these signals to specific human behaviors. Key challenges include the need for a robust, adaptable WiFi sensing framework capable of accurately interpreting complex human actions in real- time, under varying conditions, and refining feature engineering methods to adapt to environmental signal variations. These challenges are compounded by adaptability issues and the practical challenges of deploying these systems in real-world settings, alongside the need for specific hardware which hampers widespread adoption. The narrative unfolds methodically, beginning with a comprehensive review of existing challenges in WiFi-based sensing methodologies (Chapter 2), followed by the introduction of a critical Wireless-Vision Activity Recognition (WVAR) dataset to standardize HAR research and testing (Chapter 3). Chapter 4 presents WiNN, a Video-based Neural Network, as a pivotal innovation for overcoming occlusions and pushing the boundaries of HAR. This sets the foundation for Chapters 5 and 6, which delve into the creation and deployment of the spatial-temporal WiFi signal-based neural network, STWNN and the Gabor residual anti-aliasing sensing network, GraSens. These chapters underscore the evolution of HAR technology towards privacy-preserving, autonomous recognition systems capable of operating effectively in dynamic environments. In summary, this dissertation introduces a comprehensive real-time alert system which contain three innovative pipeline solutions—WiNN, STWNN, and GraSens that support swift responses and adaptations for applications like security monitoring, elderly care, and smart home automation. Especially valuable in occlusion scenarios, this system leverages WiFi signals to provide critical insights. | en_US |
