近年來由於3D列印(3D Printing)技術的成熟，使得相關設備的應用也逐漸普及。以往作為3D 列印機的資料來源的接觸式3D掃描設備，不但掃描速度過慢，有些物體(EX: 器官，精密儀器…等) 並不能以接觸的方式取得資料。本實驗室採用非接觸式的3D掃描設備，DLP(Digital Light Process)，改善了接觸式3D掃描設備的缺點。但非接觸式3D掃描設備也存在一個隱憂---無法一次掃描整個物體所有的面。非接觸式掃描設備，須將一面一面的點雲集合拼疊，在一次次的拼疊過程中，容易產生誤差。本研究致力於將非接觸式掃描設備---DLP掃描完的各個面之點雲集合，以應用K維樹的迭代最近點演算法，拼疊出一個完整的立體點雲，以便後續3D列印機使用。本研究比較了四種不同拼疊方法，迭代最近點演算法，Levenberg-Marquardt迭代最近點演算法，RANSAC Trimmed迭代最近點演算法，Levenberg-Marquardt Trimmed迭代最近點演算法。並利用四種不同模型來做不同方法的結果好壞比較(包括拼疊的速度，拼疊的誤差大小等等)。

In recent years, due to the maturity of 3D Printing technology, the application of related equipment has also been gradually popularized. In the past, as a contact-type 3D scanning device for 3D printers, not only the scanning speed was too slow, but some objects (EX: organs, precision instruments, etc.) could not obtain information with contact manner. Our lab uses non-contact 3D scanning equipment, DLP (Digital Light Process), to improve the shortcomings of contact-type 3D scanning equipment. However, there is also a hidden disadvantage with non-contact 3D scanning equipment - it is impossible to scan all the surfaces of an entire object at a time. Non-contact scanning equipment, it must align each side of the point cloud, again and again, in the process of stacking is easy to make error. This research is devoted to the collection of non-contact scanning equipment --- DLP scanned each point of the surface of the collection to apply KD Tree-Based Iterative Closest Point Algorithm, align a complete three-dimensional point cloud for the subsequent 3D column Printer use. In this study, we compare four different methods of alignment, Iterative Closest Point Algorithm, Levenberg-Marquardt Iterative Closet Point Algorithm, RANSAC Trimmed Iterative Closest Point Algorithm, and Levenberg-Marquardt Trimmed Iterative Closest Point Algorithm. And use four different models to compare the results of different methods (including stacking speed, stacking error size, etc.).

[1] H. Andresson, "Vision Aided 3D Laser Scanner Based Registration", ROC. European Conference on Mobile Robots: ECMR, pp192-197, 2007.
[2] P. Besl, N. McKay, “A Method for Registration of 3-D Shapes”, IEEE Transactions on Pattern Analysis and Machine Intelligence 14,239–256,1992
[3] G. Bradski and A. Kaehler. Learning OpenCV: Computer Vision with The OpenCV Library. O'Reilly Media, 2008.
[4] D. Chetverikov, “Fast Neighborhood Search in Planar Point Set”, Pattern Recognition Letters 12,409–412,1991
[5] D. Chetverikov, D. Stepanov, P. Krsek: “Robust Euclidean Alignment of 3D Point Sets: The Trimmed Iterative Closest Point Algorithm”, Image and Vision Computing 23, 299–309. 8, 9,2005
[6] M. Fischler, R. Bolles, "Random Sample Consensus: A Paradigm for Model Fitting Applications to Image Analysis and Automated Cartography", Proc. Image Understanding Workshop, pp. 71-88, 1980-Apr.
[7] A.W. Fitzgibbon, “Robust Registration of 2D and 3D Point Sets”, in Proceedings of The British Machine Vision Conference, 2001.
[8] M. Gross and H. Pfister, Point-Based Graphics, Morgan Kaufmann Pub, 2007.
[9] H. Kjer, J. Wilm, “Evaluation of Surface Registration Algorithms for PET Motion Correction”, Bachelor Thesis, Technical University of Denmark, 2010.
[10] M.E. Mortenson, Geometric Modeling. Wiley, New York,1985
[11] A. Nuchter, K. Lingemann, and J. Hertzberg, “Cached K-D Tree Search for ICP Algorithms,” in Proceedings of the Sixth International Conference on 3-D Digital Imaging and Modeling (3DIM), 2007.
[12] L. Piegl, W. Tiller, The NURBS Book (Monographs in Visual Communication) (Second ed.), Springer-Verlag, New York ,1997
[13] Z. Zhang, “A Flexible New Technique for Camera Calibration”,
[14] IEEE TPAMI,22(11): 1330–1334,2000
[15] https://ccjou.wordpress.com/2009/09/01/%E5%A5%87%E7%95%B0%E5%80%BC%E5%88%86%E8%A7%A3-svd/
[16] https://docs.opencv.org/2.4.13/modules/refman.html
[17] http://pointclouds.org/
[18] http://www.sciencedirect.com/science/article/pii/S092523121200728X
[19] https://zh.wikipedia.org/wiki/%E6%A2%AF%E5%BA%A6%E4%B8%8B%E9%99%8D%E6%B3%95