Supervisor’s Foreword
Abstract
Acknowledgements
Contents
Acronyms
Notation
List of Figures
List of Tables
1 Introduction
1.1 Abnormal Behaviour Detection
1.1.1 Topic Modeling
1.1.2 Change Point Detection
1.2 Key Contributions and Outline
1.3 Disseminated Results
References
2 Background
2.1 Outline of Video Processing Methods
2.1.1 Object Detection
2.1.2 Object Tracking
2.2 Anomaly Detection
2.2.1 Video Representation
2.2.2 Behaviour Model
2.2.3 Normality Measure
2.3 Topic Modeling
2.3.1 Problem Formulation
2.3.2 Inference
2.3.3 Extensions of Conventional Models
2.3.4 Dynamic Topic Models
2.3.5 Topic Modeling Applied to Video Analytics
2.4 Change Point Detection
2.4.1 Change Point Detection in Time Series Data
2.4.2 Anomaly as Change Point Detection
2.5 Summary
References
3 Proposed Learning Algorithms for Markov Clustering Topic Model
3.1 Video Representation
3.2 Model
3.2.1 Motivation
3.2.2 Model Formulation
3.3 Parameter Learning
3.3.1 Expectation-Maximisation Learning
3.3.2 Variational Inference
3.3.3 Gibbs Sampling
3.3.4 Similarities and Differences of the Learning Algorithms
3.4 Anomaly Detection
3.4.1 Abnormal Documents Detection
3.4.2 Localisation of Anomalies
3.5 Performance Validation
3.5.1 Performance Measure
3.5.2 Parameter Learning
3.5.3 Anomaly Detection
3.6 Summary
References
4 Dynamic Hierarchical Dirichlet Process
4.1 Hierarchical Dirichlet Process Topic Model
4.1.1 Chinese Restaurant Franchise
4.2 Proposed Dynamic Hierarchical Dirichlet Process Topic Model
4.3 Inference
4.3.1 Batch Collapsed Gibbs Sampling
4.3.2 Online Inference
4.4 Anomaly Detection
4.5 Experiments
4.5.1 Synthetic Data
4.5.2 Real Video Data
4.6 Summary
References
5 Change Point Detection with Gaussian Processes
5.1 Problem Formulation
5.1.1 Data Model
5.1.2 Change Point Detection Problem Formulation
5.2 Gaussian Process Change Point Detection Approach Based on Likelihood Ratio Tests
5.2.1 Likelihood Ratio Test
5.2.2 Generalised Likelihood Ratio Test
5.2.3 Discussion
5.3 Gaussian Process Online Change Point Detection Approach Based on Likelihood Estimation
5.3.1 Test Formulation
5.3.2 Theoretical Evaluation of the Test
5.3.3 Test with Estimated Hyperparameters
5.3.4 Discussion
5.4 Performance Validation on Synthetic Data
5.4.1 Data Simulated by the Proposed Generative Model
5.4.2 Data Simulated by the GP-BOCPD Model
5.5 Numerical Experiments with Real Data
5.6 Summary
References
6 Conclusions and Future Work
6.1 Summary of Methods and Contributions
6.2 Suggestions for Future Work
6.2.1 Inference in Topic Modeling
6.2.2 Alternative Dynamics in Topic Modeling
6.2.3 Gaussian Process Change Point Detection
6.2.4 Potential Applications of the Proposed Statistical Methods
References
A EM for MCTM Derivation
Appendix B VB for MCTM Derivation
Appendix C Distributions of Quadratic Forms
C.1 Quadratic form of the ``Own'' Covariance Matrix
C.2 Quadratic form of an Arbitrary Symmetric Matrix
Appendix D Proofs of the Theorems for the Proposed Test Statistic
D.1 Proof of Theorem 5.1
D.2 Proof of Theorem 5.2
Appendix E Optimisation of Gaussian Process Covariance Function Hyperparameters
References