本論文提出一個簡單且有效的方法，物件樣板為一個長方體盒子總共為六個面。先設定物件樣板每個面的平面標定點，再透過ORB(Oriented Features from Accelerated Segment Test and Rotated Binary Robust Independent Elementary Features)找尋樣板與當前畫面各自的特徵點並且描述兩者之間的特徵點，透過比對找尋出兩者相似度高的對應特徵點對關係，並利用對應點關係與隨機取樣篩選演算法來估測平面轉換齊次矩陣，如此可以透過齊次矩陣得到當前畫面平面的位置。獲取第一次的位置區域為當前畫面感興趣的範圍，排除第一次配對明顯的錯誤，降低估測齊次矩陣的誤差，獲取更好得齊次矩陣。設定相機為固定的位置並設為世界座標的參考座標點，量取每個樣板面的長方體8個角落點，距離參考座標點的實際長度當作樣板在空間中的標記點。再透過旋轉向量的方法找出當前影像畫面最適合的一個樣板平面，當前畫面位置對應空間中的標記點順序，存在一個轉換關係為空間中的變化為我們所期望未知的，只要給定4點重複去迭代計算出當前畫面的物件在空間中的位置與樣板在空間中的標記點資訊，即可求取物件需要的旋轉及平移值，在投影回來平面可知道物件的下層面位置。如此可知道追蹤長方體盒子過程的資料。

Thesis proposes a simple and effective method for a cuboid with six planes. First, set the markpoints for each side of the object, and then find the feature points of template and current image through the ORB(Oriented Features from Accelerated Segment Test and Rotated Binary Robust Independent Elementary Features) and describes the feature points each other, finding the corresponding feature point-pair relationship with high similarity through comparison, and use corresponding point relationship and random sampling algorithm to estimate the homogenous matrix of planar transformation, so it can get the current image 2-D plane position. Obtaining the first location area be the current region of interest to eliminate the obvious error of the first pairing and reduce the error of estimating the homogeneous matrix for obtaining a better homogeneous matrix. Set the camera to a fixed position and be the reference coordinates for world coordinates. Measure 8 corner points of the cuboid of each template surface, taking the actual length of the reference coordinate point as the marker point in the template space. The rotation vector method is used to find the most matching template plane for the current image, current image position corresponds to the order of points in the space, there exists a transformation relationship in which the changes in space are unknown to us, as long as iterations are repeated to calculate iteratively the position of the object in current image space and template image space with the four 2-D-3-D corresponding points, you can find the object rotation and translation values, in the projection back 2-D plane can know the position of the bottom of the cuboid. In this way, you can know the data about the process of tracking cuboid.

論文審定書 i
致謝 ii
中文摘要 iii
Abstract iv
目錄 v
圖表目錄 vii
表格目錄 x
第一章 緒論 1
1-1 研究動機與目的 1
1-2 相關研究 2
1-3 論文架構 2
第二章 研究背景 3
2-1 特徵點擷取與匹配 3
2-1-1 Oriented FAST特徵點檢測 3
2-1-2 Rotated BRIEF特徵描述子 5
2-1-3 Matching Method 9
2-2 點對關係平面轉換 10
2-2-1 Homography 10
2-2-2 RANSAC 12
2-3 相機校正 14
2-3-1 Camera Calibration 14
第三章 研究方法 22
3-1 利用ROI找尋更符合平面轉換的齊次矩陣 22
3-2 物件姿態估測方法與工作環境 24
3-2-1 估計物件在空間中姿態位置與旋轉 25
3-2-2 3D透視投影 29
3-3 角度變化量來估測換面&三維空間中物件追蹤系統 30
第四章 實驗結果 32
4-1 Homography Matrix比兩次準確度的對比性 32
4-2 所有面與當前所看到的面時間運算複雜度 35
4-3 追蹤三維物件平移與旋轉向量的趨勢與實際量測數據軌跡比較 36
4-4 影像追蹤誤差發生率 39
4-5 在複雜的環境背景上的追蹤表現 40
第五章 結論與未來展望 43
5-1 結論 43
5-2 未來展望 43
參考文獻 44

[1] Changhyun Choi, and Henrik I. Christensen, “3d textureless object detection and tracking: An edge-based approach“, In Proc. of: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3877–3884, 2012.
[2] N. Allezard, M. Dhome, and F. Jurie, "Recognition of 3D textured objects by mixing view-based and model-based representations“, Proceedings 15th International Conference on Pattern Recognition. ICPR-2000, 2000.
[3] E. Rublee, V. Rabaud, K. Konolige, and G. Bradski, “ORB: an efficient alternative to SIFT or SURF”, In Proc. of the IEEE Intl. Conf. on Computer Vision (ICCV), volume 13, 2011.
[4] D.G. Lowe, “Distinctive Image Features from Scale-invariant Keypoints”, International Journal of Computer Vision, pp.91-110, 2004.
[5] H. Bay, T. Tuytelaars, and L. V. Gool, “SURF: : Speeded Up Robust Features”, in Proc. of the 9th European Conf. on Computer Vision, Graz Austria, pp.404-417, 2006.
[6] E. Rosten, and T. Drummond, “Machine learning for high speed corner detection”, In European Conference on Computer Vision, volume 1, 2006.
[7] E. Rosten, R. Porter, and T. Drummond, “Faster and better: A machine learning approach to corner detection”, IEEE Trans. Pattern Analysis and Machine Intelligence, 32:105–119, 2010
[8] P. L. Rosin, “Measuring corner properties”, Computer Vision and Image Understanding, 73(2):291 – 307, 1999.
[9] M. Calonder, V. Lepetit, C. Strecha, and P. Fua, “Brief: Binary robust independent elementary features”, In European Conference on Computer Vision, 2010.
[10] E. Vincent, and R. Laganiere, “Detecting planar homographies ` in an image pair”, in 2nd International Symposium on Image and Signal Processing and Analysis, Pula, Croatia, pp. 182–187, June 2001.
[11] Marius Muja, and David G. Lowe, “Fast Matching of Binary Features“, 2012 Ninth Conference on Computer and Robot Vision, 2012
[12] M.A. Fischler, and R.C. Bolles, “Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography“, Communications of the ACM, 24(6):381–395, June 1981.
[13] http://en.wikipedia.org/wiki/RANSAC
[14] Z. Zhang, “A Flexible New Technique for Camera Calibration”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(11):1330-1334, 2000.
[15] Daniel Grest, Thomas Petersen, and Volker Krüger, “A Comparison of Iterative 2D-3D Pose Estimation Methods for Real-Time Applications”, 16th Scandinavian Conference on Image Analysis, 2009.
[16] Hung-Yu Tseng, Po-Chen Wu, Ming-Hsuan Yang, and Shao-Yi Chien, “Direct 3D Pose Estimation of a Planar Target”, Applications of Computer Vision(WACV), 2016 Winter Conference.
[17] H. Araujo, R. Carceroni, and C. Brown, “A Fully Projective Formulation to Improve the Accuracy of Lowe’s Pose Estimation Algorithm”, Journal of Computer Vision and Image Understanding, 70(2), 1998.
[18] Daniel Grest, “Marker-Free Human Motion Capture in Dynamic Cluttered Environments from a Single View-Point”, PhD thesis, MIP, Uni. Kiel, Kiel, Germany, 2007.