Contextual Spatial Analysis and Processing for Visual Surveillance Applications

Vikas Reddy (2011). Contextual Spatial Analysis and Processing for Visual Surveillance Applications PhD Thesis, School of Information Technol and Elec Engineering, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
s41106491_phd_finalthesis.pdf PhD thesis - Final submission application/pdf 5.37MB 8
s41106491_phd_finalthesis_abstract.pdf PhD thesis abstract - Final submission application/pdf 46.36KB 1
Author Vikas Reddy
Thesis Title Contextual Spatial Analysis and Processing for Visual Surveillance Applications
School, Centre or Institute School of Information Technol and Elec Engineering
Institution The University of Queensland
Publication date 2011-09
Thesis type PhD Thesis
Supervisor Prof Brian C. Lovell
Dr Conrad Sanderson
Total pages 159
Total colour pages 26
Total black and white pages 133
Language eng
Subjects 080104 Computer Vision
010401 Applied Statistics
170203 Knowledge Representation and Machine Learning
080106 Image Processing
Abstract/Summary Over the last few decades, computer vision has emerged as one of the predominant fields of research. Applications ranging from object detection, tracking and analysis (e.g. face recognition) have been widely explored by the research community. However, building commercial-grade systems is still a challenge, particularly in uncontrolled environments (e.g. train stations, outdoors), where the captured sequences are largely characterised by varying illumination, image noise, low resolution foreground objects, occlusions, cast shadows and non-frontal faces. In this work, we aim to address three important problems fundamental to numerous computer vision applications pertaining to object detection, tracking and analysis: (i) background estimation in cluttered sequences, (ii) foreground-background segmentation, and (iii) anomaly detection. While the first two are low-level vision problems, the last one belongs to the category of high-level vision. To address the two low-level vision tasks, most methods proposed in the literature perform analysis at pixel-level, generally not taking into account information from neighbouring pixels. The rich spatial correlations that typically exist within local regions of an image are not exploited adequately. As a result, they become sensitive to varying illumination, cast shadows, dynamic backgrounds and inherent image noise. To handle the above mentioned limitations, this work effectively utilises contextual spatial information in its analysis. Two independent algorithms employing the above principle are proposed to address the low-level vision problems of background estimation and foreground segmentation, respectively. The former problem is cast as a labelling problem in a Markov random field framework, while the latter one is viewed as a binary classification problem. The third major problem addressed by this work is detecting anomalous activities in video. The topic has drawn significant attention in recent years owing to the growing global security and safety concerns. An efficient anomaly detection technique is proposed, capable of detecting anomalies in crowded scenes, where traditional tracking based approaches tend to fail. Unlike most methods that analyse only motion based features, the proposed technique also extracts size and texture features for improved sensitivity of anomaly detection. The proposed algorithms have been designed to work in uncontrolled environments with an aim towards real-time performance. Experiments and comparative evaluations on standard datasets and real-life surveillance videos suggest that the three proposed algorithms obtain considerably better results (both qualitatively and quantitatively) than other well-known techniques available in the literature.
Keyword visual surveillance
patch analysis
background initialisation
cluttered videos
Markov Random Fields
background modelling
anomaly detection
Additional Notes colour pages: 34, 36, 42, 45, 53, 56, 60, 72-73, 75, 85, 98, 100-101, 103-105, 107-108, 119, 121, 124, 126-127, 130-131

Citation counts: Google Scholar Search Google Scholar
Created: Tue, 24 Apr 2012, 12:24:43 EST by Mr Vikas Reddy on behalf of Library - Information Access Service