Scale Adaptive Object Tracking With Diverse Ensemble /
Sara Maher El-Kerdawy
- 2015
- 100 p. ill. 21 cm.
Supervisor: Mohamed A. El-Helw
Thesis (M.A.)—Nile University, Egypt, 2015 .
"Includes bibliographical references"
Contents: Chapter 1: Introduction ................................................................................................. 11 1.1. Overview and Motivation .................................................................................... 11 1.2. Contributions........................................................................................................ 14 1.2.1. Multiple Classifier System (Diverse Ensemble) ....................................... 15 1.2.2. Scale Estimation........................................................................................ 15 1.2.3. Multiple Vehicle Detection and Tracking for Traffic Monitoring ............ 16 1.3. Publications .......................................................................................................... 17 1.4. Organization of Thesis ......................................................................................... 17 Chapter 2: State of the Art ............................................................................................ 19 2.1. Overview of Visual Object Tracking Process ...................................................... 20 2.1.1. Initialization .............................................................................................. 20 2.1.2. Appearance Modeling ............................................................................... 22 2.2. Conclusion ........................................................................................................... 31 Chapter 3: Background ................................................................................................. 33 3.1. Dimensionality Reduction ................................................................................... 33 3.1.1. Principal Component Analysais ................................................................ 35 3.1.2. Random Projections .................................................................................. 38 3.2. Multiple Classifier Systems ................................................................................. 41 3.3. Conclusion ........................................................................................................... 44 Chapter 4: An Integrated Framework for Visual Object Tracking ............................... 45 4.1. Introduction .......................................................................................................... 45 vi 4.2. Compressed Features ........................................................................................... 48 4.2.1. Filter Banks ............................................................................................... 48 4.2.2. Measurement Matrix Construction ........................................................... 48 4.3. Model Construction ............................................................................................. 49 4.3.1. Base Classifiers ......................................................................................... 50 4.3.2. Two-Step Detection .................................................................................. 53 4.3.3. Diverse Ensembles .................................................................................... 54 4.4. Conclusions .......................................................................................................... 58 Chapter 5: Results & Discussion .................................................................................. 61 5.1. Datasets ................................................................................................................ 61 5.2. Evaluation metric ................................................................................................. 62 5.3. Quantitative Analysis ........................................................................................... 62 5.3.1. Experimental setup.................................................................................... 62 5.3.2. Ensemble Evaluation ................................................................................ 63 5.3.3. Scale Adaptability Evaluation ................................................................... 67 5.4. Qualitative Analysis ............................................................................................. 67 5.5. Summary .............................................................................................................. 69 Chapter 6: Case Study: Integrated Framework for Traffic Monitoring ........................ 72 6.1. Framework architecture ....................................................................................... 72 6.2. Vehicle Detection ................................................................................................. 74 6.3. Vehicle Tracking .................................................................................................. 77 6.3.1. Tracklet Management ............................................................................... 78 6.4. Data Association .................................................................................................. 79 6.5. System Evaluation ............................................................................................... 81 6.6. Conclusion ........................................................................................................... 82 Chapter 7: Conclusions and future work ...................................................................... 87 7.1. Summary of contributions.................................................................................... 87 7.2. Future Work ......................................................................................................... 88 References .........................................................................................................................
Abstract: unrestricted environment. Even thought VOT problem is well explored in the computer vision literature, VOT remains challenging and there are lots of compromises yet to take such as robustness and time performance, adaptability and drifting. Object tracking is a crucial problem for many applications such as traffic monitoring in which we will illustrate our tracking algorithm on. Our objective of this thesis is to develop a scale and appearance adaptive object tracking algorithm using statistical machine learning techniques that also performs in real-time applications. We begin by addressing the object tracking problem by exploring two complementary components. First we explore learning appearance model using a diverse ensemble of random projections and benefiting from their sparisty yet robustness. Second we address object scaling problem to alleviate the problems introduced when the object’s size changes that differ ranging from avoiding false positive learning to updating the model with the new object appearance.