Location estimation and collective inference in indoor spaces using smartphones

De, Subham

Permalink

https://hdl.handle.net/2142/113872 Copy

Description

Title

Location estimation and collective inference in indoor spaces using smartphones

Author(s)

De, Subham

Issue Date

2021-12-01

Director of Research (if dissertation) or Advisor (if thesis)

• Sundaram, Hari
• Vasisht, Deepak

Doctoral Committee Chair(s)

• Sundaram, Hari
• Vasisht, Deepak

Committee Member(s)

• Zhai, Chengxiang
• Padmanabhan, Venkat

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Internet of Things
• Indoor Localization
• Packet reception probability
• Bluetooth Low Energy
• Contact Tracing
• Space Anonymity

Abstract

In the last decade, indoor localization-based smart, innovative services have become very popular in public spaces (retail spaces, malls, museums, and warehouses). We have state-of-art RSSI techniques to more accurate CSI techniques to infer indoor location. Since the past year, the pandemic has raised an important challenge of determining if a pair of individuals are ``social-distancing,'' separated by more than 6ft. Most solutions have used `presence'-if one device can hear another--- which is a poor proxy for distance since devices can be heard well beyond 6 ft social distancing radius and across aisles and walls. Here we ask the key question: what needs to be added to our current indoor localization solutions to deploy them towards scenarios like reliable contact tracing solutions easily. And we identified three main limitations---deployability, accuracy, and privacy. Location solutions need to deploy on ubiquitous devices like smartphones. They should be accurate under different environmental conditions. The solutions need to respect a person's privacy settings. Our main contributions are twofold -a new statistical feature for localization, Packet Reception Probability (PRP) which correlates with distance and is different from other physical measures of distance like CSI or RSSI. PRP can easily deploy on smartphones (unlike CSI) and is more accurate than RSSI. Second, we develop a crowd tool to audit the level of location surveillance in space which is the first step towards achieving privacy. Specifically, we first solve a location estimation problem with the help of infrastructure devices (mainly Bluetooth Low Energy or BLE devices). BLE has turned out to be a key contact tracing technology during the pandemic. We have identified three fundamental limitations with BLE RSSI---biased RSSI Estimates due to packet loss, mean RSSI de-correlated with distance due to high packet loss in BLE, and well-known multipath effects. We built the new localization feature, Packet Reception Probability (PRP), to solve the packet loss problem in RSSI. PRP measures the probability that a receiver successfully receives packets from the transmitter. We have shown through empirical experiments that PRP encodes distance. We also incorporated a new stack-based model of multipath in our framework. We have evaluated B-PRP in two real-world public places, an academic library setting and a real-world retail store. PRP gives significantly lower errors than RSSI. Fusion of PRP and RSSI further improves the overall localization accuracy over PRP. Next, we solved a peer-to-peer distance estimation problem that uses minimal infrastructure. Most apps like aarogya setu, bluetrace have solved peer-to-peer distances through the presence of Bluetooth Low-Energy (BLE) signals. Apps that rely on pairwise measurements like RSSI suffer from latent factors like device relative positioning on the human body, the orientation of the people carrying the devices, and the environmental multipath effect. We have proposed two solutions in this work---using known distances and collaboration to solve distances more robustly. First, if we have a few infrastructure devices installed at known locations in an environment, we can make more measurements with the devices. We can also use the known distances between the devices to constrain the unknown distances in a triangle inequality framework. Second, in an outdoor environment where we cannot install infrastructure devices, we can collaborate between people to jointly constrain many unknown distances. Finally, we solve a collaborative tracking estimation problem where people audit the properties of localization infrastructure. While people want services, they do not want to be surveilled. Further, people using an indoor location system do not know the current surveillance level. The granularity of the location information that the system collects about people depends on the nature of the infrastructure. Our system, the CrowdEstimator, provides a tool to people to harness their collective power and collect traces for inferring the level of surveillance. We further propose the insight that surveillance is not a single number, instead of a spatial map. We introduce active learning algorithms to infer all parts of the spatial map with uniform accuracy. Auditing the location infrastructure is the first step towards achieving the bigger goal of declarative privacy, where a person can specify their comfortable level of surveillance.

Graduation Semester

2021-12

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/113872

Copyright and License Information

Copyright 2021 Subham De

Owning Collections