Files in this item

FilesDescriptionFormat

application/pdf

application/pdfFREEDMAN-THESIS-2017.pdf (6MB)
(no description provided)PDF

Description

Title:LSTM and extended dead reckoning automobile route prediction using smartphone sensors
Author(s):Freedman, Ryan Taylor
Advisor(s):Gunter, Carl A
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:M.S.
Genre:Thesis
Subject(s):Dead reckoning
Long short-term memory (LSTM)
Recurrent neural network (RNN)
Neural
Network
Route
Prediction
Markov
Global positioning system (GPS)
Sensors
Accelerometer
Gyroscope
Hall effect
Spoofing
Abstract:We examine the application of two solutions to resolving automobile route shape using non-GPS sensor information provided by a smart-phone. This is motivated by the unreliability of GPS sensor information due to the ease of spoofing of GPS sensor data and areas of low signal. A trace is generated as output from this algorithm that predicts route taken. The two approaches, Extended Dead Reckoning and LSTM, are compared for their advantages and disadvantages. These concepts are explored by recording nearly one thousand miles of driving data from Virginia to Indiana to Illinois. The GPS data is used to train the LSTM neural network along with the thirty non-GPS features recorded from the smart-phone. The output is a route shape that is used to determine potential driving route and verify if a route input is correct. This method is evaluated against our implementation of dead reckoning using the same data. We find that the machine learning approach allows for precise route shape prediction with relatively constant accuracy and low sensitivity to changes in orientation. The extended dead reckoning approach proves to be substantially more accurate but sensitive to changes in orientation making the route prediction veer off if the smart-phone is moved mid-route. In a broader scope, this thesis investigates the application of a recurrent neural network (RNN) algorithm that is normally used for text-mining applications to process other types of data, namely sensor data. A more manual approach, the extended dead reckoning, is used as a comparison for this application.
Issue Date:2017-04-18
Type:Thesis
URI:http://hdl.handle.net/2142/97357
Rights Information:Copyright 2017 Ryan Freedman
Date Available in IDEALS:2017-08-10
Date Deposited:2017-05


This item appears in the following Collection(s)

Item Statistics