Date of Award

Spring 1-1-2013

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Qin Lv

Second Advisor

Richard Han

Third Advisor

Shivakant Mishra

Fourth Advisor

Li Shang

Fifth Advisor

Michael Hannigan

Abstract

People spend approximately 90% of their time indoors, and human indoor activities are strongly correlated with the rooms they are in. Room localization, which identifies the room a person or smartphone is in, provides a powerful tool for characterizing human indoor activities and helping address challenges in public health, productivity, building management, etc. Designing a room localization system that is practically useful in real-world environments is challenging. First, due to the complex multi-path propagation problem, Wi-Fi signals obtained by smartphones are dynamic and noisy. Such noise obscures the unique relationship between Wi-Fi signals and individual rooms. Second, existing room localization methods require labor-intensive manual annotation of individual rooms. The process is time-consuming and expensive, which is a key limitation of existing room localization applications. Third, knowledge of indoor floorplans is often required by room localization applications. However, indoor floorplans are either unavailable or obtaining them requires slow, tedious, and error-prone manual labor. In addition, the overhead of room localization, e.g., energy consumption, imposed on personal smartphones must be low. To tackle those challenges, this thesis proposed a set of techniques: (1) an accurate temporal n-gram augmented Bayesian room positioning method that leverages the ordered sequence information of access points and users' daily motion pattern among rooms; (2) an automatic room fingerprinting approach that identifies in-room occupancy ``hotspot(s)" using density of Wi-Fi signals, and then learns the inter-zone correlation -- thereby distinguishing different rooms; (3) an automatic floorplan construction method that determines the geometries of individual rooms, as well as room adjacency information, and then constructs an indoor floorplan through hallway layout learning and force directed dilation; and (4) an energy-efficient trip detection framework that consists of two modes: the deliberation mode learns cell-id patterns using GPS/Wi-Fi based localization methods, and the intuition mode only uses cell-ids and learned patterns for trip detection.

Share

COinS