ENHANCING ACCURACY FOR FINGERPRINT-BASED INDOOR LOCALIZATION

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "ENHANCING ACCURACY FOR FINGERPRINT-BASED INDOOR LOCALIZATION"

By

Mr. Suining HE


Abstract

The commercial potential of indoor location-based services (ILBS) has spurred 
recent development of many indoor positioning techniques. Fingerprinting has 
attracted much attention recently due to its adaptivity to none-line-of-sight 
measurement from access points (APs) and high applicability in complex indoor 
environment.

Offering quality ILBS requires accurate indoor positioning. In this thesis, we 
study several approaches to make Wi-Fi fingerprinting highly accurate. The 
approaches are to mitigate noisy signal measurement, to fuse distance sensor 
with fingerprinting, and to adaptively learn fingerprint patterns over time. We 
will conduct extensive experimental studies to validate the performance of the 
approaches.

Previous fingerprinting positioning based on certain similarity metric often 
suffers from ambiguous matching problem of reference points, resulting in high 
decision uncertainty. To address this, we propose a novel approach based on 
junction of signal tiles, which are formed based on the first two moments of 
the signals. The target location is then constrained within the junction area. 
This overcomes position ambiguity problem and achieves highly accurate 
positioning.

To further enhance the localization accuracy, we study how to fuse fingerprint 
with distance information. Our approach is applicable to a wide range of 
sensors (peer-assisted, inertial navigation sensor, beacons, etc.) and wireless 
fingerprints (Wi-Fi, Bluetooth, etc.). By a novel optimization formulation 
which jointly fuses distance bounds and measured fingerprint signals, it 
achieves low positioning errors even under complex indoor environment.

Fingerprinting accuracy deteriorates if the AP signals are altered (due to AP 
movement, partitioning, etc.). To address this, the signal map needs to be 
adapted overtime. We propose and study a novel clustering-based scheme which 
can localize targets despite AP alteration, and can identify the altered APs. 
Using a novel online learning approach, our algorithm can also adapt the 
fingerprint map to the altered signal environment.


Date:			Wednesday, 3 August 2016

Time:			2:00pm – 4:00pm

Venue:			Room 5564
 			Lifts 27/28

Chairman:		Prof. Gang WANG (CIVL)

Committee Members:	Prof. Gary Chan (Supervisor)
 			Prof. Brahim Bensaou
 			Prof. Raymond Wong
 			Prof. Wai-Ho Mow (ECE)
 			Prof. Joseph Ng (Baptist U)


**** ALL are Welcome ****