Health Monitoring Using Wearable Sensors

Abstract

Portable and low power sensors are widely used in long term remote monitoring scenarios such as health and fitness (e.g., heartbeat, body motion), agriculture, (e.g., water, soil, and climate), home environment (e.g., air quality, humidity), etc. Those sensors continuously collect critical information and transmit information to other devices (e.g., computer, cloud) for further processing and storage. For instance, a wearable sensor mobile platform enables great potential for low cost, convenient and effective remote health monitoring. Patients wearing a portable sensor wirelessly connected to their mobile phone can be monitored remotely at their home or during work without professional medical presence. However, wearable sensors suffer from various kinds of noise due to excessive activity, simplified hardware design, etc. Correlating multiple measurements to detect certain human physiological responses or status is still a difficult task. Besides, long term data monitoring requires an efficient data compression or sampling strategy to reduce the amount of data accumulated during monitoring. To address these challenges, this dissertation proposed and implemented a noise tolerant algorithm, a multi-sensor signal fusion model, and a compressive sensing based data acquisition framework aimed at providing potential solutions for the these challenges. We also proposed a framework to track sleep using wearable sensors. In addition, this dissertation also included deep learning and image processing based model to address the challenge of automatic cell nuclei detection and segmentation from microscopy images. Algorithms, models and applications implemented in this dissertation have been tested using real data from public data set repositories.