Exploration of Self-Learning Radar-based Applications for Activity Recognition and Health Monitoring
Year: 2023
Type: dissertation
Abstract: Sensor-based monitoring has proven effective in many settings for determining \npeople’s well-being and protecting their safety, even in difficult times \nlike the COVID-19 pandemic. In many applications, radio wave-based systems \nare more versatile than those based on traditional sensors, thanks to \nnon-contact sensing while preserving privacy. Health monitoring and assisted \nliving are two good examples of how such systems are finding widespread usage \nin everyday applications. Good performance in such complex monitoring \nand recognition tasks is often achieved via machine learning. In particular, \ndeep learning can aid with feature extraction, algorithm performance optimization, \nand forecasting. Yet, to learn how to tackle problems effectively, \nthe generated models usually need access to a substantial amount of data. \nFurthermore, data preparation may be time-consuming and costly, especially \nwhen handled by specialists or when required in real-time systems. Few-shot \nlearning techniques overcome these issues by adapting models to self-learn \nhow to extract meaningful information from limited data. This is feasible by \nleveraging the learning context and previously acquired knowledge. \nThis doctoral thesis is the result of research on the exploration of fewshot \nlearning techniques for radar-based applications in activity recognition \nand health monitoring. The investigation was performed by constraining the \nadaptation of radar-based solutions to limited data, ensuring the robustness \nof context generalization. The primary goal has been to investigate the use \nof limited data in very different non-contact applications, each with its own \nconstraints and requirements. Millimeter-wave radar technology and fewshot \nlearning have been used for hand gesture recognition, people counting, \nand human respiratory signal estimation. Such use cases, ranging from the \nmillimetric displacements of vital signs to the distance of moving individuals, \nrequire specific information preprocessing. The generalization learning strategy \nhas been explored for context and user adaptation while also accounting \nfor preprocessing. Some of the algorithms were adapted to run on edge devices, \nallowing for end-to-end performance estimation and adaptation. \nThe research has been carried out under a doctoral contract at the facilities \nof Infineon Technologies AG, at its headquarters in Munich, Germany.
Author: Mauro, Gianfranco
Institution: Universidad de Granada
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7028498708
