臺灣博碩士論文加值系統

將現實世界中的物體進行3D建模，對於虛擬實境及3D列印的相關應用是非常重要的。許多現今存在的3D掃描技術，藉由依序掃描的方式對靜態物體進行建模，最終皆能夠獲得令人滿意的建模結果。然而，對動態物體進行建模在許多實際應用上是高度被需要的，但是要求生物在有限的時間內保持靜止卻是不太可能的。因此，最有可能對動態物體進行建模的解決方式，是利用收集環繞物體的即時多視角資訊進行建模。
在本論文中，利用從若干視角同時拍攝物體的立體影像對，建立動態物體的三維物件模型。由於建模的物體及相機是按照給定條件所排列，物體的深度資訊可以從不同視角所拍攝的立體影像對中獲得。深度資訊是根據立體影像對間的對應像素來計算，接著轉換成與參考軸之間的徑向距離。參考軸是一條穿過物體中心並且與相機對構成的影像平面平行的軸線。隨後將徑向距離與像素在相機影像平面上的位置，轉換成該像素在三維圓柱空間中對應的體像素。在本論文中，為了使相鄰視角所估算出來的深度區域有部分重疊，每一組相機對間的角度都經過挑選。隨後將各角度的深度區域縫補在一起，以形成一個位於圓柱空間內的物體表面。在有了物體表面資訊後，物體以非重疊的立方體或立方體像素在三維空間中表示。構成物體表面的立方體像素被更進一步的平滑化成由三角平面構成物體表面。最終模型表示並儲存在常見的OBJ格式中。
在實際對一些樣本物體進行建模後，得到了普遍可接受的實驗結果，並證明了所提出的三維建模方法是可行的。

The 3D modeling of real world objects is important to the applications associated with virtual reality and 3D printing. There are many existing techniques that are able to obtain the satisfactory modeling results by scanning stationary objects in a sequential process. However, the modeling of moving objects is highly demanded in many practical applications that target on living creatures that are not likely to be stationary during a finite lapse of time. Henceforth, the modeling of moving object is most likely to be solved with the information gathered instantaneously from view angles around the object to be modeled.
In this project, the stereoscopic images of the object taken simultaneously at a number of view angles are utilized to construct the 3D model of the moving object. Given conditions on the arrangement of the object to be modeled and the cameras in the real world space, the depth information of the object can be obtained from pairs of stereoscopic images taken from different view angles. The depth information is calculated based on corresponding pixels on the pair of stereoscopic images, which is then converted to the distance away from a reference axis. The reference axis is assumed passing through the center of the object, and passing the plane parallel to the image planes of the pair of cameras. The radial distance and the position of the pixel on the camera's image plane is then converted to the three-dimensional cylindrical coordinates of a voxel corresponding to this pixel. In the project, the angle that separates each set of camera pair is so chosen such that the depth fields estimated by the associated image pair are partially overlapped. The depth fields from all angles are then stitched together to form the outer surface of the object located inside a cylindrical space. With the surface information of the object, the object is represented by non-overlapping cubes or cubic voxels in the three dimensional space. The cubic voxels that constitute the surface of the object are further smoothed into surface formed by triangular planes. The modeled object is finally represented and stored in the commonly used OBJ format. The experimental results of modeling some real sample objects are acceptable in general, and prove that the proposed 3D modeling method is feasible.

第一章 緒論 1
1.1 前言 1
1.2 研究動機 1
1.3 論文架構 2
第二章 文獻探討 3
2.1 結構光源法 3
2.2 立體視覺法 4
2.3 三維掃描技術比較 6
第三章 硬體系統 7
3.1 硬體系統設計 7
3.2 硬體系統架構 8
3.2.1 DFK AFU130-L53 CMOS彩色鏡頭 9
3.2.2 旋轉平台 11
第四章 軟體設計 12
4.1 軟體設計原理 12
4.2 圖案投影以增加影像對間的對應特徵點 13
4.3 影像前處理 13
4.4 轉換基於影像平面之深度資訊至以旋轉軸為參考中心 14
4.4.1 區塊匹配法 14
4.4.2 像差平滑化 16
4.4.3 物體深度推估 19
4.4.4 三維座標計算 22
4.6 基於旋轉軸的三維座標系統 24
4.7 表面資訊後處理 26
4.7.1 縫補全視角之重疊深度區域 26
4.7.2 使用非重疊立方體像素構成立體模型 28
4.7.4 錯誤體像素之刪除 29
4.7.3 物體表面平滑化 30
4.7.5 利用物體外表面構成三維物件模型 32
4.8 建立三維物件模型 32
4.8.1 以OBJ檔案格式表示並儲存模型 33
第五章 實驗結果 36
5.1 硬體架構展示 36
5.2 成果展示 36
5.2.1 長方體 37
5.2.2 圓球 38
5.2.3 Q版石膏像 39
5.2.4 石膏像(男) 40
5.2.5 石膏像(女) 41
5.2.6 大同寶寶 42
5.3 實驗結果分析 43
第六章 結論與未來展望 45
6.1 結論 45
6.2 未來展望 45
參考文獻 46

[1] Hsiao-Mei Chang, “3D Object Modeling from Multi-View Depth and Color Images,” M.S. thesis, Graduate Institute of Information Systems and Applications, National Tsing Hua University, January 2013.
[2] Ying Zheng, Jianglong Chang, Zhigang Zheng, and Zengfu Wang, “ 3D Face Reconstruction from Stereo: A Model Based Approach,” Proc. of IEEE International Conference on Image Processing, pp. 65-68, 2007.
[3] Hsiao-Ting Hsu, “Reconstruction of 3D Building Models based on Stereo Camera,” M.S. thesis, Graduate Institute of Electrical Engineering, National Chung Cheng University, 2015.
[4] 王正忠, 以近景攝影測量進行模型式建物重建, 成功大學測量工程學系碩士學位論文, 2002.
[5] Yoshida, T. and Fukao, T., “Dense 3D reconstruction using a rotational stereo camera,” Proc. of IEEE/SICE International Symposium on System Integration (SII),pp. 985-990, 2011.
[6] Jian-Hong Sun, Byung-Seung Jeon , and Jong-Wook Lim, “Stereo vision based 3D modeling system for mobile robot,” Proc. of IEEE International Conference on Control Automation and Systems (ICCAS), pp. 27-30, 2010.
[7] 黃盈源, 使用轉角偵測與虛擬網格法重建三維模型, 台灣師範大學機電科技學系碩士學位論文, 2011.
[8] 陳冠臻, 以影像為依據建立三維立體模型, 中山大學資訊工程系碩士學位論文, 2001.
[9] 張舜孔、邵慶旺、蔡育林、陳俊宇, 3D掃描技術應用於文化資產之適用性討論, 文化資產保存學刊, 頁63-78, 2013.
[10] 應福仁, DSP即時影像系統開發與移動目標物量測之應用, 國立台灣科技大學機械工程研究碩士學位論文, 2008.