貝蘇章臺灣大學：電信工程學研究所黃婉婷Huang, Wan-TingWan-TingHuang2007-11-272018-07-052007-11-272018-07-052005http://ntur.lib.ntu.edu.tw//handle/246246/58768在過去的十年中，多媒體的科技技術有很重大的進步與發展。 在各方面都逐漸走向數位化。從靜態的影像處理技術到影像序列，開拓了一個新的領域。因此影像序列處理技術是時空域信號處理的一個重要工具。 這篇論文的第一部份，重點放在影像序列的物體追蹤。首先，我們先介紹這一類的信號特性及頻率響應。有了這些資訊後，就能知道如何去設計物體追蹤濾波器。 接著，我們提出一個在時空域裡的一階IIR濾波器來做物體追蹤。此方法與其他用多維濾波器來追蹤物體的方法不一樣，不需要傅式轉換，完全在時空域實行。為一個能簡單設計，並且低計算複雜度的方法。不論物體軌跡是線性或是非線性，都能成功的提取出來。實驗結果可以看出成果是良好的，而且能實際即時施行。 在第二部份，將整理介紹擁有好的偵測能力的餘弦系列窗。只要在視窗裡加上可指定位置的附葉窪溝和達到高的附葉衰減率，此視窗就能把在強信號附近的弱信號偵察出來。並且強度也能夠準確的測出。此類的視窗函數可以讓使用者自行設計所需要的視窗，以達到最好的分析效果。In the past decade or so, there have been magnetic developments in multimedia technologies. It has become very clear that that all aspects of media are digitized. Extending image processing from static individual images to image sequences opens up a new world of information. Image sequence processing is becoming a tremendous tool to analysis spatio-temporal data in all areas. In part 1 of this thesis, we will focus on tracking the moving objects in the image sequence. First, the properties of this kind of signals and its frequency responses will be introduced. With this information in hand, the structure of the filter that can extract the moving object is known. Furthermore, we will propose a 1-D IIR filter in spatio-temporal domain for object tracking. Different from other multi-dimensional filter design methods for object tracking, it is easy to design and can be implemented without Fourier transform causing low computational complexity. The 1-D trajectory filter can be used to track the objects moving with linear or nonlinear trajectory without the prior knowledge of trajectories. The experimental results show that its performance is satisfactory and useful for real time practical applications. In part 2, the cosine series windows with good detectability are introduced. By adding steerable sidelobe dips and achieving high sidelobe falloff rate, the windows can be used to detect a weak signal with a strong signal nearby. Furthermore, the amplitudes of the signals are also accurately estimated. These windows can be easy to design.PRAT Ⅰ : Tracking Moving Objects in Image Sequences Using 1-D Trajectory Filters Chapter 1 Introduction 3 1.1 Problem Formulation 3 1.2 Signals in Spatial-Temporal Domain 4 Chapter 2 Related Work in Object-Tracking Methods 9 2.1 The DFT Filters 9 2.2 Three-Dimensional LDE Filters 11 2.3 Combined Discrete-Fourier-Transform and Linear-Difference-Equation (DFT/LDE) Method [2.8~2.10] 13 2.4 Motion Estimation [2.11] 14 2.5 Conclusion 15 Chapter 3 1-D Trajectory Filters for 3-D Spatio-Temporal Object Tracking 17 3.1 From 3-D to 1-D 17 3.2 Implementation of 1-D Trajectory Filter in Spatial-Temporal Domain 22 3.3 Experimental Results 27 3.3.1 The Linear Trajectory Case 27 3.3.2 The Nonlinear Trajectory Case 30 3.4 Comparison of Trajectory Filters with DFT, Combined DFT/LDE and LDE Filters 32 3.4.1 Assumptions 32 3.4.2 Comparison 33 3.5 Conclusion 35 Chapter 4 Object Tracking By Optical Flow Estimation Using Structure Tensor Method 37 4.1 Optical Flow 37 4.2 Structure Tensor Method for Optical Flow Estimation 38 4.3 The Minors of the Structure Tensor 39 4.4 General Model for Multiple Motions 41 4.5 Solutions for n Transparent Motions 42 4.5.1 Separation of the Motion Vectors 44 4.5.2 Confidence Measures 45 4.5.3 Hierarchical Algorithm for Multiple Motions 45 4.6 Experimental Results 46 4.6.1 Motion Estimation of a Synthetic Sequence 46 4.6.2 Motion Estimation of a Real Sequence 46 4.7 1-D Trajectory Filters Using Optical Flow Estimation 48 4-8 Conclusion 50 Chapter 5 Conclusions and Future Work 53 PRAT Ⅱ: Cosine Series Windows with High Sidelobe Falloff Rate and Steerable Sidelobe Dips Chapter 6 Introduction 57 Chapter 7 Detectability of Cosine Series Windows 59 7.1 The Cosine Series Windows 59 7.1.1 Class ⅠWindow：Minimum High-order Sidelobe Amplitude 60 7.1.2 Class Ⅱ Window：Minimum Main-lobe Width 61 7.2 Cosine Series Windows with High Sidelobe Falloff Rate and Steerable Sidelobe Dips 62 7.2.1 The Improvement of Class Ⅰ Windows 63 7.2.2 The Improvement of Class Ⅱ Windows 68 7.3 Experimental Result 73 7.4 Conclution 76 Reference 791772942 bytesapplication/pdfen-US軌跡估測軌跡濾波器餘絃視窗偵測能力trajectory estimationtrajectory filterscosine series windowsdetectbilitythesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/58768/1/ntu-94-R92942105-1.pdf