臺灣博碩士論文加值系統

早期因傳統攝影測量於影像上要進行觀測量提取不易，甚至傾斜過大而無法從影像中取得量度值，所以並未針對傾斜攝影技術(Oblique Photograph)做過多的著墨，然而隨著多視立體視覺 (Multi View Stereo, MVS) 三維重建技術開始被應用，使其無人載具 (Unmanned Aerial Vehicles, UAV) 於飛行限制放寬，並搭載消費型相機於UAV上進行拍攝，不但減少拍攝成本，解析度及解算穩定性相較於航空攝影高，因此本研究利用現代立體視覺的方法進行處理流程中包含相機率定、三維重建、正射影像糾正及透過單張影像進行量測。
在相機自率定(Self-Calibration)技術開始發展前，多是使用預率定方式求解出內外方位參數，預率定所使用的方式能夠求得穩定解，但計算所需條件複雜繁瑣並費時，自率定的發展改善了預率定方法解算條件複雜及費時的問題，但因自率定方法存在著不穩定因素，因此為了比較率定方法適用性，實驗選定高應大校區進行拍攝，並針對不同控制點數量探討率定成果。
現今商用傾斜攝影處理系統多用傳統攝影測量空中三角方法進行三維重建，而所使用的相機也是專業型航拍相機，畫素雖高但造價昂貴，因此本研究設計使用無人飛行載具搭載自組傾斜相機進行拍攝，後續再利用現代立體視覺的SfM方法進行三維重建，並生產DEM模型提供後續單像量測使用，實驗分為校區及台中都會區進行重建，並針對模型完整度及重建精度進行探討。
為了能夠達到影像快速應用，本研究利用共線式原理對傾斜影像單片量測，主要因傾斜影像側面資訊豐富並能充分表現出標的物的特徵結構，實驗區選定校區進行建物的量測，並探討點雲、單像量測、地測之間的差異。
相機率定實驗結果顯示兩種方法無顯著的差異，甚至部分自率定成果優於預率定，因此得知自率定方法可達到快速及方便的成果；由三維重建結果得知傾斜攝影因大量多於觀測關係精度優於垂直攝影，並且能夠完成較美觀之模型；單像量測能夠快速獲取量測資訊，並且能過取得2pixel以內的較差，因此透過直接對影像進行量度方法能夠提供災害應變或是場勘後一個參考依據。

It is not easy to measure statistics on images through traditional photography in the early period, it is even unable to measure the figures from images, resulted in overbank. Thus, there is no further description about oblique photograph. Nevertheless, since MVS( Multi View Stereo) 3D reconstruction starts being applied, it loosens the aviation requirements of UAV. Combining the digital compact camera with UAV( Unmanned Aerial Vehicles) to shooting, it can not only reduce the shooting cost, but the resolution and stability are much higher than aero photography by comparison.Therefore the study includes camera calibration, 3D reconstruction, generats the true Orthophoto based on morden view stereo processing.
Before camera self-calibration begin to develop, people use pre-calibration to solve the intrinsic parameters and exterior orientation parameters, pre-calibration could get the steady answer, but the requirements are complex and time-consuming. The development of self-calibration improve the problems of pre-calibration, however, the requirements of self-calibration is unstable, so there are accuracy analysis of the camera self-calibration, therefore, to compare with the applicability of camera calibration methods, the experiment select region of National Kaohsiung University of Applied Sciences and choose different numbers of control points to explore the results of the calibration.
Today commercial Oblique Photograph processing system and 3D reconstruction are using traditional photogrammetric aerial triangulation method by the perfectional camera with high resolution but it is also expensive, so this study design uses digital camera by unmanned aerial vehicle to shoot and use modern methods of stereo vision SfM reconstruction, mainly discusses tilt integrity model photography and 3D reconstructio accuracy, and provide DEM let measurement based on single image of oblique photograph to use.The experiment selects National Kaohsiung University of Applied Sciences and Taichung city to do reconstruction and explore completion and accuracy of 3D reconstruction model.
In order to make the images apply fleetly, this research utilize collinearity principle to measure single image of oblique photograph, mainly because the lateral information of tilted images are rich, and it can fully display the characteristics and construction of the object. The experiment selects a region of National Kaohsiung University of Applied Sciences to compare the difference of dense cloud, measurement based on single image and survey.
The part of the camera calibration show there is no significant difference between the two methods, self-calibration even better than pre-calibration. From the experiment of 3D-reconstruction show because of a lot of observation, accuracy of oblique photography method is better than traditional photography, even get complete model. Measurement of Single image gets 3D information quickly and have less than 2 pixel difference, therefore, through direct measurement method can provide images for disaster response or after a field survey reference.

摘要 i
Abstract iii
誌謝 v
目錄 vii
圖目錄 xi
表目錄 xv
第一章 緒論 1
1.1 研究背景目的 1
1.2 三維重建基於傳統攝影測量之侷限性 1
1.3 現代與傳統攝影測量之差異 2
1.4 傾斜攝影介紹 2
1.4.1 傾斜攝影特點 3
1.4.2 傾斜攝影成像幾何 3
1.4.3 傾斜攝影應用範圍 4
1.5 研究方法 5
1.6 預期成果 8
1.7 論文架構 9
第二章 文獻回顧 11
2.1 攝影測量 11
2.2 相機率定 11
2.2.1 預率定 12
2.2.2 自率定 12
2.3 多視立體視覺 12
2.3.1 從相機運動恢復場景結構(Structure from motion, SfM) 13
2.4 傾斜攝影技術 13
2.4.1 傾斜相機介紹 14
2.4.2 傾斜影像處理系統 15
2.5 曲面重建 16
2.5.1 離散建模 16
2.6 紋理貼圖(Texture Mapping) 17
第三章 SfM-MVS三維重建基本理論 19
3.1 相機率定 19
3.1.1 相機預率定 22
3.1.2 率定標樣式 24
3.1.3 Kruppa方程式的相機自率定 26
3.1.4 絕對二次曲線的自率定 28
3.2 基於SfM三維重建 28
3.2.1 影像匹配 29
3.2.2 SfM演算法(Structure from Motion) 37
3.3 稠密點雲重建 39
3.3.1 PMVS演算法 40
3.3.2 面片模型 41
3.3.3 光度差異函數 41
3.3.4 圖片模型 42
3.3.5 面片重建 43
3.4 曲面重建及紋理貼圖 44
3.4.1 八叉樹空間點分類(Octree) 45
3.4.2 Delaunay三角網格化 45
3.5 紋理貼圖(Texure Mapping) 46
第四章 傾斜攝影成像幾何與資料處理 47
4.1 傾斜攝影成像幾何 47
4.1.1 尺度解析度不均勻 47
4.1.2 遮擋嚴重(單一物件的遮擋及物件 - 物件的遮擋) 48
4.2 傾斜影像管理與檢索 50
4.3 傾斜攝影特徵匹配適用性 51
4.3.1 ASIFT演算法原理 52
4.3.2 ASIFT處理步驟 52
4.3.3 AIF演算法原理 53
4.3.4 AIF處理步驟 53
4.4 基於傾斜攝影的面片重建 - 改進PMVS方法 54
4.5 傾斜影像單像量測 55
4.5.1 求取單張影像坐標流程 56
4.5.2 傾斜影像外方位參數解算 57
4.5.3 基於共線式進行單張影像量測 59
第五章 實驗成果與分析 61
5.1 實驗一 61
5.1.1 實驗資訊 62
5.1.2 率定實驗步驟及流程 64
5.2 實驗二 81
5.2.1 實驗資訊 81
5.2.2 實驗流程 83
5.2.3 高應大傾斜攝影重建精度及建模成果 83
5.2.4 台中都會區建模成果 89
5.2.5 傾斜攝影於不同航帶之比較 91
5.3 實驗三 93
5.3.1 實驗資訊 93
5.3.2 單像量測實驗流程 94
5.3.3 實驗成果 94
第六章 結論與展望 99
6.1 結論 99
6.2 展望 101
第七章 參考文獻 103
附錄一 相機率定成果 107
1.1 DJI相機率定值 107
1.2 QX1_01 相機率定值 119
1.3 QX1_02 相機率定值 131
1.4 QX1_03 相機率定值 143

[1]Börner A., Test results obtained with the LH Systems ADS40 Airborne Digital Sensor. International Archives of Photogrammetry and Remote Sensing, Vol. 33, Part B4, Amsterdam, Netherlands, 2000, pp.871-878.
[2]Gerke M. and Kerle N. Automatic Structure Seismic Damage Assessment with Airborne Oblique Pictomerty(c) Imagery [OL], 2012.
[3]Wang Y. D., Schultz S., Giuffrida F., Pictomerty’s Proprietary Airborne Digital Imaging System and Its Application in 3D City Modeling [OL], 2012.
[4]Karb O. N., Schroth R. Oblique Aerial Photography: a status review [OL], 2012.
[5]Grenzdorffer G.J., Guretzki M. Friedlander I., Photogrammetric Image Acquisition and Image Analysis of Oblique Imagery. The Photogrammetric Record, 23(124):372-386, 2008.
[6]McClone J.C .e, Mikhail E.M.e., Bethel J.S.e. (Editors), Manual of photogrammetry. American Society for Photogrammetry and Remote Sensing (ASPRS), Bethesda, 1151 pp.Snavely, 2004.
[7]Höhle J., Photogrammetric Measurements in Oblique Aerial Images. Photogrammetric Fernekundung Geoinformation, 2008(1), pp. 7-14.
[8]Stephen Schultz, Frank Giuffrida, Robert Gary, Mosaic Oblique Image and Methods of Making and Using Same, 2008, United States:US2008/0123994A1.
[9]Schultz Stephen L., Giuffrida, Frank D., Gary, Robert L., Mondello, Charles, Method and Apparatus for Capturing, Geolocating and Measuring Oblique Images, 2006, Europe:06012675.2.
[10]Faig W., Calibration of Close-range Photogrammetric System: Mathematical Formulation. Photogrammetric Eng. Remote Sensing. 1975, 41(12), pp. 1479-1486.
[11]Brown D. C., Decentering Distortion of Lenses. Photogrammetric Eng. Remote Sensing. 1966, pp. 444~462.
[12]Abdel-Aziz Y. Y., and Karara H. M., Direct Linear Transformation into Object Space Coordinates in Close-Range Photogrammetry. In Proc. Close-Range Photogrammetry, Jan. 1971, pp. 1-18.
[13]McGlone J., Mikhail E., Bethel J., Mullen R., Manual of Photogrammetry, 2004.
[14]Luhmann T., Robson S., Kyle S., Harley, S., Close Range Photogrammetry: Principles, 2006.
[15]Slam S., Manual of Photogrammetry, 1980.
[16]Tsai R. Y., An Efficient and Accurate Camera Calibration Technique for 3D Machine Vision, 1986, pp.364-374.
[17]Huang T. S., Tsai R. Y., Three Dimensional Motion Estimation from Image –Space Shift. Proc. of IEEE Int. Conf. on ASSP. Piscataway：IEEE, 1981, vol2, pp. 1136-1139.
[18]Hartley R., Self-Calibration of Stationary Cameras. Int. Journal of Computer, 1997, 22(1), pp.5-23.
[19]Hartley R. Euclidean Reconstruction and Invariants from Multiple Images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1994, 16(10), pp. 1036-1041
[20]Heyden, A., Astrom, K. Euclidean reconstruction from constant intrinsic parameters. In: Proceedings of International Conference of Pattern Recognition, Vienna, 1996, pp.339-343.
[21]Marr D., A Computational Investigation into The Human Representation and Processing of Visual information, San Francisco, USA: W. H. Freeman, 1982, pp.86-156.
[22]Martinec D., Pajdla T., Robust Rotation and Translation Estimation in Multiview Reconstruction, IEEE Conference on Computer Vision and Pattern Recognition, Minneapolis, USA, 2007, pp.1-8.
[23]Brown M., Lowe D. G., Unsupervised 3D Object Recognition and Reconstruction in Unordered Datasets, Proceedings of the International Conference on 3D Digital Imaging and Modeling, Ottawa, Canada, 2005, pp.56-63.
[24]Schaffalitzky F., Zisserman A., Multi-View Matching for Unordered Image Sets, or How do I Organize my Holiday Snaps?, Proceedings of the 7th European Conference on Computer Vision, Copenhagen, Denmark, 2002, pp.414-431.
[25]Vergauwen M., Van Gool L., Web-Based 3D Reconstruction Service, Machine Vision and Applications, 2006, 17(6), pp.411-426.
[26]Snavely N., Seitz S. M., Szeliski R., Modeling the World from Internet Photo Collections, International Journal of Computer Vision, 2008, 80(2), pp.189-210.
[27]Tomasi C., Kanade T., Shape and Motion from Image Streams Under Orthography: a Factorization Method, International Journal of Computer Vision, 1992, 9(2), pp. 137-154.
[28]Maybank S. J., Faugeras O. D., A Theory of Self-Calibration of a Moving Camera, Int Journal of Computer, 1992 , 8(2), pp.123-151.
[29]Faugeras O., What Can Be Seen in Three Dimensions with an Uncalibrated Stereo rig, Computer Vision-ECCV'92,Lecture Notes in Computer Science, Vo1.588, Springer Verlag, pp.563-578, 1992.
[30]Pollefeys M., Van Gool L., M. Vergauwen, et al. Visual Modeling with a Hand-Held Camera, International Journal of Computer Vision, 2004, 59(3), pp.207-232
[31]Kerle N. R., Stekelenburg F., Near-real Time Post-Disaster Damage Assessment Withairborne Oblique Video Data, Geo-Information for Disastermanagement Gi4DM, Proceedings of the 1stinternationalsymposium on Geo- Information for Disaster Management, TheNetherlands, 21-23 March, Berlin, Springer, 2005, pp. 337–353.
[32]Mitomi H., Yamazaki F., Matsuako M., Automated Detection of Building Damage due to Recent Earthquakes Using Aerial Television Image. Proceedings of the 21st Asian Conference on Remote Sensing. Asian Conference on Remote Sensing, Taiwan: Taipei, 2000, pp. 1-7.
[33]Zwicker M., Pfister H., Van baar J., Gross M. H., EWA Volume Splatting. In proceedings of IEEE Visualization 2001. IEEE Computer Society Press, 2001,pp. 29-36.
[34]Zwicker M., Rasanen Jussi., Perspective Accurate Splatting. In proceedings of the Eurographics 2004, Computer Graphics Forum, Conference Issue, 2004,pp.461-470.
[35]Fergusion J.C. Multivariable Curve Interpolation, Report D2-22504, The Boeing Co. Seattle, Washington, 1963.
[36]Bajaj C. L., Bermardini F., Xu G., Automatic Reconstruction of Surfaces and Scalar Fields from 3D Scan. In Cook R ed. Proceeding of SIGGRAPH’95. Danvers: Assison –Wssley Publishing Company, 1995, pp.109-118.
[37]Yang M., Lee E., Segmentation of Measured Point Data Using a Parametric Quadric Surface Approximation, Computer Aided Design, 1999, 31, pp.499-457.
[38]Zwcker M., Pfister H., Van Barr J., Gross M. H., EWA Volume Splatting. In Proceedings of IEEE Visualization 2001. IEEE Computer Society Press, 2001, pp. 29-36.
[39]Zwicker M., Rasanen J., Perspective Accurate Splatting. In proceedings of the Eurographics 2004. Computer Graphics Forum, Conference Issue, 2004, pp.461-470.
[40]Hoppe H. Surface Reconstruction from Unorganized Point . Washington : Computer Science and Engineering, University of Washington, 1994, pp.13-29.
[41]Amenta N., Bern M., Kamvyssclis M. A New Voronoi Based Surface Reconstruction Algorithm. In proceeding of Siggarph98, Orlando, ACMpress, 1998, pp.415-420.
[42]Delaunay B., Sur la sphère vide, Izvestia Akademii Nauk, SSSR, Otdelenie Matematicheskikh i Estestvennykh Nauk, 1934, (7), pp.793–800.
[43]戴嘉境，基於多幅圖像的三維重建理論集算法研究，上海交通大學碩士論文，2012。
[44]李怡蓁，基於多視立體視覺的三維重建，國立高雄應用科技大學碩士論文，2013。
[45]宋碧波，傾斜航空實景影像系統構建關鍵技術研究，河南里工大學碩士論文，2012。
[46]陳俊君，立體視覺三維重建影像匹配方法的研究，國立高雄應用科技大學碩士論文，2014。
[47]Furukawa Y., Ponce J., Accurate, Dense, and Robust Multiview Stereopsis, IEEE Transactions on pattern analysis and machine, 2010, 32(8), pp.1362-1376.
[48]張庭榮，SIFT演算法於立體對影像匹配與影像檢索應用之研究，國立高雄應用科技大學碩士論文，2008。
[49]楚萬秀，近景攝影測量標誌點設計及資訊處理研究，東華大學碩士論文，2008。
[50]李琳慧，基於自適應物方面元的無人機多事影像密集演算法，武漢大學碩士論文，2011。
[51]Moreal J. M., Yu G., ASIFT: A New Framework for Fully Affine Invariant Image Comparison. SIAM Journal on Imaging Sciences，2009，2(2), pp.438-469．
[52]肖武雄，基於特徵不變的傾斜影像匹配算法研究與應用，西安科技大學碩士論文，2014。
[53]Debevec P. E., Taylor C. J., and Malik J., Modeling and Rendering Architecture from Photographs”, ACM SIGGRAPH, 1996.
[54]Christian F., Russell S., Avideh Z., Automated Texture Mapping of 3D City Models With Oblique Aerial Imagery, Proceedings of the 2nd International Symposium on 3D Data Processing, 2004, pp. 396-403.
[55]Ebner, H., Self Calibrating Block Adjustment, Bildmessung und Luftbildwessen, 1976, 44, pp.128-139.