Date of Award:

5-2020

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Electrical and Computer Engineering

Committee

Rose Qingyang Hu

Committee

Jacob Gunther

Committee

Yan Sun

Abstract

Human action recognition and monitoring are a significant part of a smart healthcare system, which allows us to remotely monitor the behavior of patients or elderly persons to record their daily activities and to ensure safety. Ubiquitous smart wearable sensors are becoming convenient for generating and transmitting those healthcare data for further processing. Interpretation of such huge amount of data requires advanced learning systems where most important information can be extracted and analyzed. For example, current classification algorithms are not robust enough to differentiate among numerous variations of human actions due to the lack of sufficient processed samples. Thus, one aspect of this thesis focuses on how to incorporate raw data with the processed one in order to build abetter classification framework. Furthermore, for a remote surveillance system, excessive amount of data need to be transmitted to the receiver end, which is not only an energy-intensive process but also may jeopardize the real-time processing capability because of the overwhelming number of packets received and numerous packet losses. This challenge is addressed by the second part of the thesis, which designs an efficient framework to reduce the amount of insignificant information for transmission. The proposed scheme can not only reduce the number of packets but is also robust to packet loss.

Checksum

53a7dc53e20028a2b77a9f714a3ff765

Available for download on Thursday, May 01, 2025

Share

COinS