Attentional Action-Driven Deep Networks for Visual Object Tracking

다국어 초록 (Multilingual Abstract)

This dissertation proposes a novel visual tracking method which is controlled by sequential actions learned by deep reinforcement learning.
In the recent trackers using deep networks, tracking-by-detection scheme is adopted to select the target position with the highest matching score.
The tracking-by-detection scheme achieves a good performance in a simple manner but is inefficient in exploring candidates.
We propose an efficient action-driven deep tracker which is controlled by sequential actions trained by deep reinforcement learning.
In contrast to the existing trackers using deep networks, the proposed tracker is designed to achieve a light computation as well as satisfactory tracking accuracy in both location and scale.
The deep network to control actions is pre-trained using various training sequences and fine-tuned during tracking for online adaptation to target and background changes.
The pre-training is done by utilizing deep reinforcement learning as well as supervised learning.
The use of reinforcement learning enables even partially labeled data to be successfully utilized for semi-supervised learning.
In addition, this dissertation tackles a tracking problem of an object interacting with other objects in a complex scene such as basketball game scenes containing various interactions among players and painting motions.
For this purpose, we design a multi-agent architecture diverse interaction movements among neighboring objects near the target object.
The multi-agent architecture is designed so that the main tracker could determine a proper action by utilizing the states of neighboring trackers.
Through extensive evaluation, the proposed tracker is validated to achieve a competitive performance that is much faster than state-of-the-art deep network-based trackers.

목차 (Table of Contents)

• Chapter 1 Introduction 1
• 1.1 Background 1
• 1.2 Related Work 4
• 1.2.1 Visual Object Tracking 4
• 1.2.2 Action-Driven Approach 6
• 1.2.3 Deep Reinforcement Learning 7
• 1.2.4 Multi-agent Reinforcement Learning 8
• 1.3 Contents of Research 9
• 1.4 Thesis Organization 12
• Chapter 2 Preliminaries 13
• 2.1 Reinforcement Learning 13
• 2.1.1 Markov Decision Process 14
• 2.1.2 Reinforcement Learning Problem 15
• 2.2 Policy Gradient 16
• Chapter 3 Action-Driven Visual Tracking 23
• 3.1 Overview 23
• 3.2 Problem Settings 25
• 3.3 Action-Decision Network Architecture 28
• 3.4 Training Methods for Action-Decision Network 32
• 3.4.1 Training ADNet with Supervised Learning 32
• 3.4.2 Training ADNet with Reinforcement Learning 35
• 3.5 Online Adaptation in Tracking 40
• 3.6 Implementation Details 42
• 3.6.1 Pretraining the ADNet 42
• 3.6.2 Online Adaptation of the ADNet 42
• Chapter 4 Interacted Action-Driven Visual Tracking 47
• 4.1 Overview 47
• 4.2 Problem Settings 50
• 4.3 Proposed Method 51
• 4.3.1 Baseline 51
• 4.3.2 Multi-agent Architecture 55
• 4.4 Training Methods for Multi-agent ADNet 58
• 4.4.1 Training Multi-agent ADNet with Supervised Learning 58
• 4.4.2 Training of Multi-agent ADNet with Reinforcement Learning 60
• 4.5 Online Adaptation in Tracking 60
• 4.6 Implementation Details 61
• 4.6.1 Pretraining 61
• 4.6.2 Online Adaptation 61
• Chapter 5 Experiments 65
• 5.1 Action-Driven Visual Tracking 65
• 5.1.1 Datasets 67
• 5.1.2 Self-comparison 69
• 5.1.3 Analysis 69
• 5.1.4 State-of-the-art Comparison 76
• 5.1.5 Qualitative Results 79
• 5.2 Interacted Action-Driven Tracking 85
• 5.2.1 Datasets 85
• 5.2.2 Self Comparison 88
• 5.2.3 Quantitative Results 88
• 5.2.4 Qualitative Results 89
• Chapter 6 Conclusion 95
• 6.1 Concluding Remarks 95
• 6.2 Future Works 96