林巍聳臺灣大學：電機工程學研究所張鈺偉Zhang, Yu-WeiYu-WeiZhang2007-11-262018-07-062007-11-262018-07-062004http://ntur.lib.ntu.edu.tw//handle/246246/53483用立體機器視覺擔任行動機器人的運動導向時，需由不同視角的影像求取立體的空間資訊，而影像對正可以把左右影像轉換成對正的影像，使每一對水平掃描線都與共極線吻合，如此一來，在對正過的左右影像中搜尋對應點時，只需在相同水平的掃描線中搜尋即可，搜尋對應點的工作因而變成比較有效率的一維演算。要達成影像對正，至少必須知道左右影像上的八對對應點，才能求解對正的轉換矩陣，前人的研究是用人工挑選出這八對對應點，本研究則發展出一套自動化的彩色樣本對應法來找到足夠的對應點，彩色樣本用顏色、亮度和形狀來區別，對應點即取為所估測對應的彩色樣本的形心，這套自動化的對正方法已經用兩組立體攝影機的實驗結果證實為有效而且精準。Using machine stereovision for the guidance of mobile robot needs to extract the 3D information from images captured at different viewpoints. The image rectification can transform the associated images into a format, which have the horizontal image lines coinciding with the epipolar lines. Thus the stereo matching can search in the rectified images simply along epipolar lines, which is basically one-dimensional. Rectification of uncalibrated images needs to know no less than eight pairs of corresponding points to estimate the homographies. Conventional approach presented in literature provides the corresponding points by manual selections. This research develops an automated method of matching color patterns to find the required number of corresponding points. A pair of corresponding points is estimated as the centroid of a pair of corresponding color patterns. The patterns are discriminated by according to their color, intensity and shape. This automated approach is shown to be efficient and accurate. Experimental results from two sets of binocular camera are shown.Contents Chapter 1 Introduction 1 1-1 Bacground 1 1-2 Review on rectification 3 1-3 Motivation and contribution 4 1-4 Organization 5 Chapter 2 Binocular Stereovision 6 2-1 Introduction to the camera calibration and 3-D reconstruction 6 2-1-1 Homogeneous coordinates 6 2-1-2 Camera model 7 2-1-3 Camera calibration 10 2-1-4 3-D reconstruction 11 2-2 The image rectification 12 2-2-1 Epipolar geometry 12 2-2-2 Epipolar line constraint 12 2-2-3 Fundamental matrix 13 2-2-4 Reconstructing depths from the rectified images 15 2-3 Comparing the rectified with un-rectified methods 18 Chapter 3 Rectification of un-calibrated images 20 3-1 The theorem of rectification 20 3-2 The forms of the rectifying transformation 22 3-2-1 Mapping the epipole to infinity 22 3-2-2 The matching transformations 25 3-2-3 The rectifying homographies 26 3-3 The rectification algorithm 27 Chapter 4 Automatic Process for Finding The Corresponding Points 30 4-1 The color patterns 30 4-2 Finding the centroids of color patterns 31 4-3 The corresponding points 48 4-4 Six color patterns 49 4-5 Detecting the wrong matching pairs 52 4-5-1 The Least Median of Squares estimator 53 4-5-2 Experimental results 56 Chapter 5 Experimental Results and Discussions 59 5-1 Automated image rectification 59 5-1-1 Color patterns of nine squares 60 5-1-2 Color patterns of six squares 69 5-2 Stereo matching 75 5-2-1 Block matching 75 5-2-2 Experiment results 77 5-3 The depth reconstruction 87 5-4 Applications of the automated image rectification 89 Chapter 6 Conclusions and Future Works 91 References 93 Acknowledgement I1629835 bytesapplication/pdfen-US立體機器視覺對應點影像對正共極線立體攝影機epipolar linesstereovisionrectificationhomographiesuncalibratedthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/53483/1/ntu-93-R91921056-1.pdf