傅楸善臺灣大學：資訊工程學研究所許毓承Hsu, Yu-ChengYu-ChengHsu2007-11-262018-07-052007-11-262018-07-052007http://ntur.lib.ntu.edu.tw//handle/246246/53919在日常視訊會議中，使用者彼此之間沒有視線接觸的問題一直存在。我們在使用電腦做視訊會議的時候，眼睛是看著螢幕中心的對方的。但是因為一些硬體上的限制，我們沒有辦法將攝影機架設在螢幕的正中心。這樣造成了使用者感覺對方總是沒有看著他。如果兩方有適當的眼神接觸，使用者會覺得比較受尊重以及更專心的參與這場會議。 我們這篇論文打算提出一個可以從螢幕左右兩邊的相機輸入影像並產生高品質虛擬正面影像的系統。而為了適應大部分的環境，而不用只能在乾淨的背景前面使用此系統，我們也採用了去除背景的方式，解決一些瑕疵的問題也將影像遮蔽的問題也考慮進去。In daily video conferencing, eye contact problem always exists because we can not set a camera in the middle of the screen. Users will feel more respected and pay more attention to this conferencing with proper eye contact. This thesis proposes a high quality virtual view synthesis system with two video streams from two cameras. Cameras were located in left and right sides of the monitor. To adapt to most environments (background), this thesis also proposes a background subtraction method. Artifacts and occlusion problems are also considered.口試委員會審定書 i 誌 謝 ii 摘 要 iii Abstract iv 目 錄 v 圖 錄 vi Chapter 1 Introduction 1 1.1 Problem Definition 1 1.2 Motivation 2 Chapter 2 Related Work 3 2.1 Model-Based and Image-Based Approaches 3 Chapter 3 System Configuration 8 3.1 Experimental Environment 8 3.2 System Pipeline 10 3.2.1 Camera Calibration and Image Rectification (Offline Processing) 11 3.2.2 Background Model Estimation 14 3.2.3 Background Subtraction 15 3.2.4 Scanline Matching 16 3.2.5 Find Minimum Cost Path and Project It to Virtual Image Plane (Image Blending) 16 Chapter 4 Implementation Details 19 4.1 Background Model Estimation and Background Subtraction 19 4.2 Scanline Matching 23 4.2.1 Corner Matching and Color Matching 23 4.3 Find Minimum Cost Path and Project It to Virtual Image Plane (Image Blending) 26 Chapter 5 Results 28 Chapter 6 Conclusions and Future Works 35 Reference 362150749 bytesapplication/pdfen-US注視方向矯正動態規劃視訊會議背景去除eye-gaze correctiondynamic programmingvideo conferencebackground subtractionthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53919/1/ntu-96-J93922008-1.pdf