傅立成臺灣大學：電機工程學研究所賴傳文Lai, Chuan-WenChuan-WenLai2007-11-262018-07-062007-11-262018-07-062006http://ntur.lib.ntu.edu.tw//handle/246246/53558在影像追蹤的應用當中，多目標物追蹤系統(MTT System)常會遇到一個問題，即移動的目標物無可避免地會在影像當中彼此交錯，因此對個別目標物作外觀的量測過程中便產生了目標物間的相依性。在本論文當中，我們提出了使用結合影像相似度(Joint Image Likelihood)以及對目標物距離相機的相對深度層次(Depth Level)的假設來分辨交錯中的目標物，實驗證明即使目標物的外觀相似仍可維持目標追蹤。為了同時達到多目標物的偵測及追蹤，我們將SIR粒子濾波器(Particle Filter)延伸為分散式SIR粒子濾波器；然而為了增加系統在處理交錯目標物的分辨時的效能，我們採用以馬可夫鏈蒙地卡羅(Markov Chain Monte Carlo)為基礎之粒子濾波器，此方法可以有效率地解決目標物交錯時的高維度狀態估測。本論文另一個重點在於提出主動相機的控制策略，使該系統可自動移動相機視角使得監視範圍中能包含最多的資訊量。最後，透過實驗來驗證此即時系統整體的效能及可靠性。In visual tracking, multi-target tracking (MTT) systems encounter the problem that unavoidably moving targets may occlude each other and the measurement process of each target becomes dependent. We construct a tracking system with considering joint image likelihood to recognize targets, even though the appearances of the target are identical. Also, the multiple hypotheses of the targets’ depth level are utilized for occlusion handling. In order to enhance system performance, we extend the sampling importance resampling (SIR) particle filter with the separated importance functions for tracking each target and detection. Furthermore, when targets occlude together, the state vector of these targets is transferred into a joint state vector, and the MCMC (Markov Chain Monte Carlo) based particle filter is then proposed for efficient sampling in the high-dimensional joint state during occlusion.　Furthermore, a control strategy for the active camera is proposed in order to move the camera such that the surveillance area will contain the most information. The overall performance is validated in the experiments and shows the robustness with real-time tracking.摘要 I Abstract II Table of Contents III List of Figures V Chapter 1 Introduction 1 1.1. Motivation 1 1.2. Brief Literature Review 2 1.3. Contribution 4 1.4. Thesis Organization 5 Chapter 2 Preliminaries 7 2.1. Bayesian Filter 7 2.1.1. Bayes’ Theorem 7 2.1.2. Generic Stochastic Filter from Bayesian Perspective 8 2.1.3. Kalman Filter 12 2.2. Particle Filter 16 2.2.1. Monte Carlo Integration 17 2.2.2. Sequential Importance Sampling (SIS) Particle Filter 18 2.2.3. Resampling 20 2.2.4. Bootstrap/Sampling Importance Resampling (SIR) Particle Filter 22 2.3. Gradient Ascent 23 2.3.1. Kernel Density Estimation 24 2.3.2. Mean Shift Algorithm 27 Chapter 3 Modified Multi-Target Tracking Algorithm 33 3.1. Single-Target Tracking 33 3.2. JVPDA Filter for Multi-Target Tracking 37 3.2.1. Review of Conventional JVPDA Filter 38 3.2.2. Image Feature Simlarity 44 3.3. Visual Tracking with Occlusion Handling 45 3.4. Multi-Target Tracking Based on Modified Particle Filter 49 3.4.1. Importance Sampling with Separated Importance Functions 51 3.5. Occlusion Handling 52 3.5.1. Estimation Based on Joint Image Likelihood 53 3.5.2. Multiple Hypotheses with Temporal Consideration 54 3.6. MCMC-based Particle Filter 58 3.6.1. Impoverishment Phenomenon in SIRPF 59 3.6.2. Properties of Markov Chains 60 3.6.3. Markov Chains in Discrete State Space 62 3.6.4. Metropolis-Hastings Algorithm 64 3.6.5. Definition of Ergodicity 66 3.6.6. MCMC-based Particle Filter Algorithm 68 Chapter 4 Implementation 71 4.1. Measurements 71 4.1.1. Outline Likelihood 71 4.1.2. Appearance Likelihood 72 4.1.3. Color Distribution Likelihood 73 4.1.4. Outer Color Likelihood 74 4.2. Implementation of Single-Target Tracking System 75 4.3. Implementation of Multi-Target Tracking System 76 4.4. Active Camera Control 80 4.4.1. Coordinate Transformation 80 4.4.2. Control Strategy 82 Chapter 5 Experiment Results 85 5.1. Results of Single-Target Tracking System 85 5.2. Results of the MMT System using Modified Particle Filter 85 5.3. Results of MMT system using MCMC-based particle filter 89 5.4. Results of Active Camera 92 Chapter 6 Conclusion and Future Work 95 6.1. Conclusion 95 6.2. Future Work 96 References 973665966 bytesapplication/pdfen-US多目標物追蹤影像追蹤馬可夫鏈蒙地卡羅粒子濾波器結合影像相似度主動相機Multi-Target TrackingVisual TrackingMarkov Chain Monte CarloParticle FilterSequential Monte CarloJoint Image LikelihoodActive Camerathesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53558/1/ntu-95-R93921010-1.pdf