臺灣博碩士論文加值系統

透過三維重建的技術將地形完整的呈現出來在電腦視覺的領域中一直以來都是相當重要的一個研究主題，因為透過重建後的三維模型，我們可以得到物體更多且更精確的資訊。可惜目前立體對應問題對電腦而言仍舊是一個眾所皆知的難題，且尚未出現強健有效率的立體對應計算方法。雖然電腦無法正確地找出立體影像間的對應關係，但是若給定正確的立體對應點，電腦即可計算出精確的三維座標。另一方面，人類先天就對立體對應擁有很好的處理能力，但是卻無法準確地說出物體的距離。為了結合人類與電腦視覺系統的長處，許多研究學者開發了互動式的地形三維重建系統，經由使用者所提供的概略立體對應關係來輔助電腦進行立體對應的計算， 以減少計算時間。但是互動式系統最重要的就是要有一個高效率的人機介面，在地形重建的應用中，傳統的滑鼠與鍵盤應該不是最有效率的介面。因此在本論文中，我們開發了一套互動式的地形三維重建系統，這套系統包含了雙眼視線追蹤系統、立體視覺產生裝置和地形三維重建計算三個部份。透過雙眼視線追蹤系統針對要重建的區域進行視線掃描，就可以得到該區域概略的影像對應關係，再利用這個對應關係來輔助電腦將不可能的區域剔除，以達到減少運算量而縮短計算時間的目的，最後將影像間完整的對應關係透過三維重建的技術重建出地形原始的三維座標位置。

Reconstructing terrain models has been an important research topic in the field of computer vision. However, stereo matching is still an open question and hence there is no robust enough stereo correspondence algorithm available. Although computers can not solve the stereo matching problem correctly, given correct stereo correspondences, the 3-D coordinates can be computed very accurately. On the other hand, humans can handle stereo matching problem efficiently, but can not show the exact distance of the object. Therefore, in order to take advantages of the human vision system and the machine vision algorithms, many researchers try to develop interactive 3-D terrain model reconstruction systems.
The 3-D terrain model can be efficiently computed with the information provided by a human operator.
However, the most important issue in an interactive system is the efficiency the human-computer interface.
It is obvious that the traditional mouse and keyboard interfaces are not suitable for interactive terrain reconstruction.
Therefore, in this thesis, we develop an interactive 3-D terrain model reconstruction system which consists of a binocular gaze tracker,
a stereo display, and a 3-D reconstruction algorithm.
Rough stereo correspondences can be obtained by using the recorded binocular gaze tracking results.
Then, the rough correspondence map can be used to assist computer eliminating the irrational matching regions.
With this interactive system, the total processing time can be reduced effectively and the 3-D terrain model can be correctly reconstructed.

目錄
1 簡介 1
1.1 動機與目的 . . . . . . . . . . . . . . . . . . . . . . 1
1.2 文獻探討 . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 研究成果 . . . . .. . . . . . . . . . . . . . . . . . . 5
1.4 章節架構 . . . . . . . . . . . . . . . . . . . . . . . 5
2 系統硬體架構與軟體流程 7
3 立體視覺與雙眼視線追蹤系統 11
3.1 立體視覺 . . . . . . . . . . . . . . . . . . . . . . 11
3.1.1 共軛幾何 (Epipolar Geometry) . . . . . . . . . . . 12
3.1.2 輸入影像 . . . . . . . . . . . . . . . . . . . . . 15
3.2 雙眼視線追蹤系統 . . . . . . . . . . . . . . . .. . . . . 17
3.2.1 瞳孔影像追蹤流程 . . . . . . . . . . . . . . . . . . . 17
3.2.2 系統校正 . . . . . . . . . . . . . . . . . . . . . 17
4 立體匹配與三維重建 20
4.1 自我校正 . . . . . . . . . . . . . . . . . . . . . . . 20
4.1.1 RANSAC 強健估測法 . . . . . . . . . . . . . . . . . . . 21
4.1.2 Non-linear estimation . . . . . . . . . . . . . . . . 22
4.2 輸入影像正規化. . . . . . . . . . . . . . . . . . . . . . 22
4.2.1 計算投影轉換矩陣. . . . . . . . . . . . . . . . . . . . 23
4.2.2 影像重新採樣. . . . . . . . . . . . . . . . . . . . . 27
4.3 影像對應點計算 . . . . . . . . . . . . . . . . . . . 29
4.3.1 最大流量等於最小切割定理 . . . . . . . . . . . . . . . . 30
4.3.2 以最大流量等於最小切割定理進行對應點計算. . . . . . . . . 35
4.4 三維重建. . . . . . . . . . . . . . . . . . . . . . . . 43
5實驗結果與討論 49
5.1系統介面. . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 影像正規化成果. . . . . . . . . . . . . . . . . . . 52
5.3 立體對應點計算結果. . . . . . . . . . . . . . . . . . . . 55
5.3.1 使用灰階影像進行對應點計算. . . . . . . . . . . . . . . 55
5.3.2 使用 Y、Cr、Cb 影像計算對應點 . . . . . . . . . . . . . 59
5.3.3 系統操作時間 . . . . . . . . . . . . . . . . . . . . . 65
6 結論與未來展望 66

[1] H. Thomas, E. Charles, L. Ronald, and C. Stein, 蕭世文 譯, “演算法導論,Intruduction to Algorithms,” 2003.
[2] P. Sturm and S. Maybank, “On plane-based camera calibration: A general al-
gorithm, singularities, applications,” Computer Society Conference on Computer
Vision and Pattern Recognition, pp. 432–437, 1999.
[3] C. Loop and Z. Zhang, “Computing rectifying homographies for stereo vision,”
Computer Society Conference on Computer Vision and Pattern Recognition, vol. 1,
pp. 125–131, 1999.
[4] R. Hartley, R. Gupta, and T. Chang, “Stereo from uncalibrated cameras,” Pro-
ceeding s Computer Vision and Pattern Recognition conference, 1992.
[5] R. Hartley, “Estimation of relative camera positions for uncalibrated cameras,”
European Conf. On Computer Vision, pp. 579–587, 1992.
[6] M. Brown, D. Burschka, and G. Hager, “Advances in computational stereo,”
PAMI, vol. 25, no. 8, pp. 993–1008, 2003.
[7] P. Aschwanden andW. G. uhl, “Experimental results from a comparative study on
correlation-type registration algorithms,” Robust Computer Vision, pp. 268–289,
1992.
[8] R. Zabih and J. Woodfill, “Non-parametric local transforms for computing vi-
sual correspondence,” Proceedings of European Conference on Computer Vision,
pp. 151–158, 1994.
[9] A. Gruen, “Adaptive least squares correlation: a powerful image matching tech-
nique,” South African Journal of Photogrammetry, pp. 175–187, 1985.
[10] V. Kluth, G. Kunkel, and U. Rauhala, “Global least squares matching,” Interna-
tional Geoscience and Remote Sensing Symposium, vol. 2, pp. 1615–1618, 1992.
[11] F. Bignone, O. Henricsson, P. Fua, and M. Stricker, “Automatic extraction of
generic house roofs from high resolution aerial imagery,” European Conference on
Computer Vision, pp. 85–96, 1996.
[12] V. Venkateswar and R. Chellappa, “Hierarchical stereo and motion correspondence
using feature groupings,” Int’l J. Computer Vision, vol. 15, pp. 245–269, 1995.
[13] C. Schmid and A. Zisserman, “The geometry and matching of curves in multiple
views,” European Conference on Computer Vision, pp. 104–118, 1998.
[14] C. Tomasi and R. Manduchi, “Stereo matching as a nearest-neighbor prob-
lem,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20,
pp. 333–340, 1998.
[15] S. Roy and I. Cox, “A maximum-flow formulation of the n-camera stereo corre-
spondence problem,” International Conference on Computer Vision, pp. 492–499,
1998.
[16] S. Roy, “Stereo without epipolar lines: A maximum-flow formulation,” Interna-
tional Journal of Computer Vision, pp. 147–161, 1999.
[17] H. Zhao, “Global optimal surface from stereo,” International Conference on Pat-
tern Recognition, vol. 1, pp. 101–104, 2000.
[18] I. Thomos, S. Malasiotis, and M. Strintzis, “Optimized block based disparity es-
timation in stereo systems using a maximum-flow approach,” Proceedings of the
International Symposium on Computer Graphics, Image Processing, and Vision,
1998.
[19] S. Sinha and M. Pollefeys, “Multi-view reconstruction using photo-consistency and
exact silhouette constraints: A maximum-flow formulation,” IEEE International
Conference Computer Vision, 2005.
[20] J. Chang, K. Lee, and S. Lee, “Stereo matching using iterated graph cuts and
mean shift filtering,” Asian Conference on Computer Vision, pp. 31–40, 2006.
[21] R. Hartley and A. Zisserman, “Multiple view geometry in computer vision,” 2000.
[22] A. Fitzgibbon, M. Pilu, and R. Fisher, “Direct least squares fitting of ellipses,”
International Conference on Pattern Recognition, vol. 1, pp. 253–257, 1996.