本研究的目的是以模板法來修補3D掃描人體表面資料之破洞，現今的3D人體計測可以在短短的幾十秒內良測出詳細精確的3D人體體型資料。不過量測所得到的第一手原始資料並不完整，仍有許多殘缺的破洞。導致破洞的主要原因為遮蔽效應與邊陲斜射效應，這是3D人體計測的一大缺憾。先前修補這些破洞的方法主要為，自建網格修補法及縫合貼補法。這兩種方法都是非常的耗時費力，同時也產生極大的失真。
為了改善上述的問題，本研究將嘗試應用模板法修補破洞，由於原始資料的破洞以腋下、跨下以及左手右手最為嚴重，因此我們就以它們當作例子來說明。。模板法的概念，可假設一個女孩子的腿上有許多疤痕，這些疤痕就如同原始掃描資料上面的破洞，將褲襪視為模板，在穿上褲襪之後，褲襪會完全貼合腿部的表面，因此疤痕就被褲襪覆蓋了。本研究以模特兒人台，為男性及女性分別建立了4個模板：腋下模板，跨下模板以及左手、右手模板。每個模板上皆指定了20個至37個不等的特徵點。相同的，在原始資料上也指定了相同數目的對應控制點。模板法就將模板上的特徵點，利用仿射轉換原理投影到原始資料的控制點上，同時也將模板上的所有網格全部投影到原始資料上，於是破洞就被修補了。模板法比起先前的自建網格修補法及縫合貼補法，快速而且傳真。

The purpose of this study is to fill holes in 3D scanned body surface data using template matching method. Modern 3D human body scanners can measure the shape and size of human body surface with high speed and great accuracy. Nevertheless these scanners usually can not measure the entire body surface completely, so the resulting 3D original data consist of many holes. The fundamental causes of holes are missing views and grazing angle views. These holes are significant drawbacks of 3D anthropometry. To fill these holes, in the past, two techniques were usually used, namely planar triangulation method and stitching method. These two methods are all very time-consuming and usually exhibit significant distortion.
To overcome these problems, this study demonstrate a procedure of filling holes using template matching method, and because the most severe holes in the original data are in the armpit, the crotch and the hand, so they are used as descriptive examples. The template matching is analogue to that women’s legs consist of many scares, the scares are considered as holes, by warring a pant hose stocking, the stocking is shaped and fitted perfectly on the woman’s legs, the holes (scares) were completely covered with stocking. For each gender, four regional templates, the armpit, the crotch, the right hand and the left hand were contracted from display model manikins. All these templates are designated with from 20 to 37 featured points. Similarly, on the original data the same numbers of corresponding control points were identified. Then the template matching method employs affine transformation algorithms to project all the featured points of the template onto the control points of the original data, and them all the meshes on the template as well, as a result the holes on the original data were filled. This method is much preferable than the planar triangulation method and stitching method both in speed and fidelity.

[1] NEW METHODS FOR SURFACE RECONSTRUCTION FROM RANGE IMAGES, 1997,Brian Lee Curless

[2] P. Besl and N. McKay. A Method for Registration of 3-D Shapes. IEEE Trans. Pattern Analysis and Machine Intelligence, 14:239–256, 1992.

[3] The space of human body shapes:reconstruction and parameterization from range scans, ACM SIGGRAPH 2003, 27-31 July 2003, San Diego, CA, USA 2003 ACM Brett Allen, Brian Curless, Zoran Popovi´

[4] T. Pajdla, L. V. Gool, Matching of 3-D Curves using Semi-differential
Invariants, in: 5th International Conference on Computer Vision, 1995, pp. 390–
395.

[5] S. Rusinkiewicz, M. Levoy, Efficient Variants of the ICP Algorithm, in: Third
International Conference on 3D Digital Imaging and Modeling, 2001.

[6] Robust Euclidean alignment of 3D point sets:the Trimmed Iterative Closest Point
algorithm, Volume 23, Number 3, March 2005, Image and Vision Computing, Dmitry Chetverikov, Dmitry Stepanov, Pavel Krsek

[7]樣板模型對應外嵌模組使用者手冊，工研院

[8]Linear algebra and its application, David C.LAY, Addison-Wesley