This decommissioned ERA site remains active temporarily to support our final migration steps to https://ualberta.scholaris.ca, ERA's new home. All new collections and items, including Spring 2025 theses, are at that site. For assistance, please contact erahelp@ualberta.ca.
- 312 views
- 363 downloads
Low-rank and Sparse based Representation Methods with the Application of Moving Object Detection
-
- Author / Creator
- Shakeri, Seyed Moein
-
In this thesis, we study the problem of detecting moving objects from an image sequence using low-rank and sparse representation concepts. The identification of changing or moving areas in the field of view of a camera is a fundamental step in visual surveillance, smart environments, and video retrieval. Recent methods based on low-rank representation have been successfully employed for change detection; however, they still have difficulties in handling the following situations. First, the existing methods rely on a batch formulation whose computational complexity grows with the size of the input data. Secondly, they are not able to deal with significant illumination change including shadow and abrupt or discontinuous change in illumination. This thesis proposes solutions to the above two problems, with the end goal of developing a reliable low-rank and sparse decomposition to perform an efficient and accurate change detection method in such cases, especially for the moving object detection tasks.
To cope with the computational complexity of the low-rank methods for change detection, we propose a sequential solution using contiguous sparsity constraint. We formulate the problem of moving object detection under integration of online robust PCA and low-rank matrix approximation with contiguous sparse outliers. This combination enables us to extract foreground objects in the case of online and long-term continuous tasks, which cannot be achieved by the batch formulation.
To deal with discontinuous change in illumination, we first propose a robust representation of images against illumination, which can be used in classification, place recognition, and change detection applications. We then build a prior map from the invariant representation, and formulate a low-rank and invariant sparse decomposition (LISD) method by incorporating the original representations and the obtained prior maps. This joint framework empowers the accuracy of object detection by separating the sparse outliers into real changes and illumination change matrices. We also propose an iterative version of LISD (ILISD) to improve the performance of LISD by updating the prior map. Experiments on challenging benchmark datasets demonstrate the superior performance of the proposed method under complex illumination changes.
As the second solution to deal with discontinuous change in illumination and to boost the accuracy of foreground detection, we propose a robust solution based on the multilinear (tensor) data low-rank and sparse decomposition framework. In this method we first introduce a way to provide multiple invariant representations of an image as priors that can characterize the changes in the image sequence due to illumination. To deal with concurrent, two types of changes, we employ two regularization terms, one for detecting moving objects and the other for accounting for illumination changes, in a novel unified framework named tensor low-rank and invariant sparse decomposition (TLISD). Extensive experiments on challenging datasets demonstrate a remarkable ability of the proposed formulation to detect moving objects under discontinuous change in illumination.
-
- Graduation date
- Fall 2019
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.