Efficient and effective automatic surveillance approaches /
Visual surveillance networks are installed in many sensitive places in the present world. The London Underground and London Heathrow Airport have more than 5000 cameras for the purposes of physical security, for example. Human security officers are required to continuously stare at large numbers of...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur:
Kulliyyah of Information and Communication Technology, International Islamic University Malaysia,
2013
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/5351 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Visual surveillance networks are installed in many sensitive places in the present world. The London Underground and London Heathrow Airport have more than 5000 cameras for the purposes of physical security, for example. Human security officers are required to continuously stare at large numbers of monitors simultaneously, and for lengths of time at a stretch. Constant alert vigilance for hours on end is difficult to maintain for human beings with natural cognitive limits on concentration and attention spans. Studies have shown that the optimal concentration span for a human being is about 25-30 minutes. It is thus important to remove the onus of detecting unwanted activity, events which rarely occur, from the human security officer to an automated system. This is the objective of this thesis. The literature survey has revealed that while many researchers have proposed solutions to this problem in the recent past, significant gaps remain in existing knowledge. First, most existing algorithms involve high complexities, and need significant memory and storage resources. Second, no quantitative performance analysis/comparison is provided by most researchers. Third, most commercial systems require specific-purpose hardware and sophisticated equipment. This thesis proposes algorithms where the computational, storage and memory complexities are independent of time, making the algorithms naturally suited to online use. In addition, the proposed methods have been shown to work with the simplest, run-of-the-mill surveillance systems that may already be publicly deployed. Furthermore, direct quantitative comparisons of detection performances between algorithms have been provided by means of Receiver Operating Characteristics (ROC) curves. More specifically, algorithms from the field of machine learning have been selected to address the stated problem. The algorithms selected are adaptive, portable, affordable and practical, and have run times of the order of hundredths of a second. The effectiveness of each proposed algorithm has been tested through application to real image streams from two complementary data sources. The test data sets include both outdoor and indoor environment, both centralized and distributed architectures, and both low-resolution and high-resolution cameras. The proposed algorithms have also been compared with representative benchmark algorithms presently in existence for this task, and quantitative performance metrics have been computed for each algorithm. The quantitative values for the detection and false alarm rates demonstrate that the proposed algorithms provide superior performance relative to the benchmark algorithms, and also yield high detection rates with low false-alarm rates on an absolute scale, on both data sets. Furthermore, the proposed algorithms meet the actual runtime limitations enforced by a real-time application such as this. |
|---|---|
| Physical Description: | xiv, 128 leaves : ill. ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 119-126). |