黃漢邦臺灣大學：機械工程學研究所張景富Chang, Ching-FuChing-FuChang2007-11-282018-06-282007-11-282018-06-282006http://ntur.lib.ntu.edu.tw//handle/246246/61264本文的主要目的為發展適用於雙眼立體視覺系統的三維影像重建理論。我們提出一個新的方法來解決立體比對中最困難的對應點問題，並加入一些幾何的限制來減少模零兩可的對應。經過雙向驗證的對應點比對更能確保在估測視差時的正確性。將稠密的視差圖經過後處理之後使其更為平滑可靠，再利用立體三角量測使之回復三維景物的特徵。 當攝影機的擺設趨近於幾何限制，我們採用投射中值精確的校正以減少錯誤比對的發生。此外，藉由採用一色彩轉換模擬人類視覺系統，處理立體比對中色彩資訊不豐富之區域，結果證實模擬人類視覺色彩認知結合所提出的比對方法能有效地偵測深度與重建三維場景。所提出的比對方法，除了能有效的減少運算量以達到近即時之外，在影像深度不連續之區域的比對更有明顯的改善。In this thesis, a binocular stereo vision system is constructed for depth estimation and 3D image reconstruction. We propose a new method called “color weighted correspondence algorithm” to solve correspondence problems. Geometric constraints are added to reduce the ambiguities in stereo matching. Left-right consistency check is used to increase the confidence for matching results. Post-processing helps remove the outliers and make original dense disparity map smooth and reliable. The three-dimensional characteristics can be reconstructed by stereo triangulation. Projected median is used to rectify cameras before performing the stereo matching in case mismatches happen. It performs well as the cameras are set up in approximately correct position. The YCC color model is adopted to simulate human vision system to deal with textureless regions. It is then applied to the “color weighted correspondence algorithm” to determine the correspondence. It validates the feasibility of humanoid color perception with stereo matching methodology to estimate depth and reconstruct 3D scene.摘要 I Development of a Near Real-Time Stereo Vision System II Abstract II Contents III List of Tables V List of Figures VI Chapter 1 Introduction 1 1.1 Motivation 1 1.2 Related Works 2 1.2.1 Static Stereo 3 1.2.2 Dynamic Stereo 4 1.2.3 Active Stereo 4 1.2.4 Stereo Using More Than Two Images 5 1.3 Objectives and Contributions 6 1.4 Thesis Organization 7 Chapter 2 Background Knowledge 9 2.1 Computational Stereo 9 2.2 Central Projection and Epipolar Geometry 13 2.2.1 Central Projection and Pinhole Camera Model 13 2.2.2 Epipolar Geometry 20 2.3 Camera Calibration 21 2.4 Stereo Matching 24 2.4.1 Local Correspondence Methods 25 2.4.2 Global Correspondence Methods 27 2.5 Fast Block Matching Algorithm 28 2.5.1 Concept of Winner-Update Strategy 29 2.5.2 Winner-Update Algorithm for Block Matching 31 Chapter 3 Disparity Estimation 34 3.1 Overview of Methodology 34 3.2 Color Transformation 36 3.3 Camera Rectification by Using Projected Median 38 3.4 Correspondence Determination 41 3.4.1 Constraints 41 3.4.2 Color Weighted Correspondence Algorithm 44 3.4.3 Consistency Check 48 3.5 Disparity Estimation and Post-Processing 50 3.6 Stereo Triangulation 51 Chapter 4 Experimental Results 55 4.1 Test for Proposed Algorithm 55 4.1.1 Test for Color Transformation 55 4.1.2 Test for Color Weighted Correspondence Algorithm 61 4.2 Performance Evaluation 63 Chapter 5 Stereo Vision System 72 5.1 System Overview 72 5.2 Camera rectification 74 5.3 Stereo Camera Calibration 75 5.4 Depth Estimation 80 Chapter 6 Conclusions 82 6.1 Conclusions 82 6.2 Future Works 83 References 853058470 bytesapplication/pdfen-US立體視覺stereo visionthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/61264/1/ntu-95-R93522823-1.pdf