University of Twente Student Theses

Login

Enhancing Human Gesture Recognition: An unsupervised Approach using Wi-Fi Channel State Information

Vadlamani, Shalini (2024) Enhancing Human Gesture Recognition: An unsupervised Approach using Wi-Fi Channel State Information.

[img] PDF
1MB
Abstract:This paper explore the utilization of Model-Agnostic Meta-Learning (MAML) for the purpose of recognizing human activities. It specifically focuses on leveraging Channel State Information (CSI) data obtained from Wi-Fi signals. We assessed our customized MAML implementation in diverse and demanding situations, encompassing variations in people, locations, orientations, and environmental conditions. The results exhibited substantial enhancements in accuracy and resilience, attaining nearly flawless performance in numerous instances. The training accuracy steadily increased across the epochs, demonstrating the model's ability to successfully adjust to various users, locations, orientations, and environmental circumstances. The validation accuracy exhibited comparable patterns, so affirming the model's ability to generalize. The graphs of training and validation loss showed effective learning and quick convergence, especially when using a learning rate of 0.001. The model's outstanding performance was confirmed by the confusion matrices, which showed average accuracies of 99.96% for person variety, 98.47% for location variation, 97.46% for orientation sensitivity, and 99.68% for environmental changes. This study emphasizes the capacity of MAML to improve human activity recognition in various real-world environments, providing useful knowledge for future progress in the field.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Embedded Systems MSc (60331)
Link to this item:https://purl.utwente.nl/essays/101920
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page