Login (DCU Staff Only)
Login (DCU Staff Only)

DORAS | DCU Research Repository

Explore open access research and scholarly works from DCU

Advanced Search

Visual object detection from lifelogs using visual non-lifelog data

Ye, TengQi (2018) Visual object detection from lifelogs using visual non-lifelog data. PhD thesis, Dublin City University.

Abstract
Limited by the challenge of insufficient training data, research into lifelog analysis, especially visual lifelogging, has not progressed as fast as expected. To advance research on object detection on visual lifelogs, this thesis builds a deep learning model to enhance visual lifelogs by utilizing other sources of visual (non-lifelog) data which is more readily available. By theoretical analysis and empirical validation, the first step of the thesis identifies the close connection and relation between lifelog images and non-lifelog images. Following that, the second phase employs a domain-adversarial convolutional neural network to trans- fer knowledge from the domain of visual non-lifelog data to the domain of visual lifelogs. In the end, the third section of this work considers the task of visual object detection of lifelog, which could be easily extended to other related lifelog tasks. One intended outcome of the study, on a theoretical level of lifelog research, is to iden- tify the relationship between visual non-lifelog data and visual lifelog data from the perspective of computer vision. On a practical point of view, a second intended outcome of the research is to demonstrate how to apply domain adaptation to enhance learning on visual lifelogs by transferring knowledge from visual non-lifelogs. Specifically, the thesis utilizes variants of convolutional neural networks. Furthermore, a third intended outcome contributes to the release of the corresponding visual non-lifelog dataset which corresponds to an existing visual lifelog one. Finally, another output from this research is the suggestion that visual object detection from lifelogs could be seamlessly used in other tasks on visual lifelogging.
Metadata
Item Type:Thesis (PhD)
Date of Award:January 2018
Refereed:No
Supervisor(s):Gurrin, Cathal and Smeaton, Alan F.
Subjects:Computer Science > Lifelog
Computer Science > Machine learning
DCU Faculties and Centres:Research Initiatives and Centres > INSIGHT Centre for Data Analytics
DCU Faculties and Schools > Faculty of Engineering and Computing > School of Computing
Use License:This item is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 3.0 License. View License
Funders:Irish Research Council for Science Engineering and Technology. Grant Number GOIPG/2013/330
ID Code:22193
Deposited On:05 Apr 2018 10:07 by Cathal Gurrin . Last Modified 18 Dec 2019 12:23
Documents

Full text available as:

[thumbnail of TENGQI_THESIS.pdf]
Preview
PDF - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
11MB
Downloads

Downloads

Downloads per month over past year

Archive Staff Only: edit this record