Date of Award

Spring 5-2018

Embargo Period

5-21-2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

Advisor(s)

Pei Zhang

Second Advisor

Hae Young Noh

Abstract

With the growth of networked smart devices in indoor environments, human information acquisition becomes essential for these devices to make the environment smart and people’s lives more convenient. These networked systems, which are often referred to as Cyber-Physical Systems (CPS), learn and make decisions collaboratively based on data input. The data could come from sensors that perceive various signals in the physical world, human input, etc. In this thesis, I will focus on information acquisition based on data from sensing the physical world. The major challenges to accurately interpreting the information these systems perceive result from the complexity of the physical world. An extreme solution to this problem is to have a large number of sensors or sensing configurations that collect a large amount of data. Ideally, we could then have labeled data for each sensing condition and possible scenario in order to accurately model the world. However, in the real world, such solutions could be difficult if not impossible to achieve due to constraints on the hardware, computational power, and (labeled) dataset. This thesis targets this problem and sets the goal of obtaining accurate indoor human information through limited system configurations and limited labeled data. A new concept of utilizing structures as sensors is presented as the foundation of the system. The intuition is that people induce ambient structures to vibrate all the time, and their activities and information can be inferred from this vibration. To achieve that with the aforementioned constraints, an understanding of the physical world (that has been studied for centuries in multiple disciplines) is used to assist the sensing and learning process for more accurate information acquisition from sensor data.

Share

COinS