【作者】 欧阳文斌；

【导师】 唐莉萍；

【作者基本信息】 东华大学 ， 通信与信息系统， 2010， 硕士

【摘要】 随着我国计算机技术和社会经济的发展,人们对服装的质量、合体性、个性化的要求越来越高,现有的二维服装CAD技术已经不能满足纺织服装业的CAD应用要求,服装CAD迫切需要由目前的平面设计发展到立体三维设计。其中,如何根据模特的照片序列重建出模特的三维模型将具有重要的研究意义。本课题来源于国家重点学科“服装设计与工程”服装CAD与美化服饰等方面的研究,在应用计算机图像图形处理、CAD、人工智能、图形学与显示等方面的新技术、新成果的基础上,研究三维对象重构这一新兴、交叉领域中关键性的理论、应用问题及其有效解决方法。三维对象重构的方法有许多种,由于视觉外壳的重构方法速度快,且只需要知道重构对象的轮廓,所以本文将在研究中将此种方法作为讨论基础。通过对彩色图像序列的二值化操作,提取对象轮廓。利用镜面反射多角度几何关系计算相机内外参数。最后借助OpenGL函数库开发模型浏览器,对重构模型进行可视化验证。本文就物体轮廓提取提出了一系列解决方案。通过对常用彩色图像处理时使用的色彩空间的介绍,以及颜色空间结合颜色相似度处理的结果进行对比从而确定颜色空间的选择。针对以往颜色相似度计算方法的不足,提出了新的颜色相似度的计算方法。根据颜色相似度定义,以及对典型差影法的详细研究,提出了基于颜色相似度的差影法。通过分析二维Otsu动态阈值算法,给出了适合本文二值化方法——基于颜色相似度的动态阈值算法。针对二值化处理结果,提出了对噪声处理的方案：采用数学形态学分离了与对象连接的噪声；内轮廓填充修补了对象内部存在的空洞；区域填充算法则以面积为阈值对图像整体进行了噪声的消除。通过比较各种相机标定方法,本文结合基于镜面拍摄场景方法,提出了利用成像图像各个对象之间的几何关系推导出相机内外部参数,并且通过视角的交点的一致性确定了图像中对象轮廓顺序。这些都为三维对象的模型重构起到了决定性的作用。重构后的对象模型以网格(Mesh)数据结构储存,保留了对象点在空间的相对位置关系,以及点与点的三角关系。借助OpenGL函数库对重构模型进行旋转,平移,放大等可视化操作。同时验证了重构模型的准确性。更多还原

【Abstract】 With computer technology and socio-economic development, apparel’s quality, fit, and individuality for human are needed strongly requirements. The extant Two-Dimensional apparel CAD technology has been unable to meet the textile and apparel industry CAD application requirements. CAD technology for apparel should evolve from thecurrent Two-Dimensional graphic design to theThree-Dimensional design. Thus, to reconstruct the 3D model from 2D images will be a significantresearch interest. This project is originated from the national key disciplines-Apparel Design and Project which is researched on apparel CAD and garment decoration. Based on the image processing, CAD, artificial intelligence, and computer graphics technology, this project is researched on 3D object reconstruction.Visual Hull algorithm is chosen from many 3D reconstruction methods, because of its higher algorithm efficiency and less parameter’s requirements (only needs the object’s contours). Thus, this research contents are found on Visual Hull algorithm. In order to extract objects’ contours, color images are needed binarization. Camera’s inner and exterior parameters are calculated from geometrical relation which is formed by mirrors’reflection. An OpenGL model browser is developed to test and verify the reconstructed model.This paper proposed a series of contour extraction solutions. Color similarity processing results in different color spaces are determined the final color space. A new color similarity algorithm is presented to nake up the former algorithm’s deficiency. According to the definition of color similarity and the typical method of background subtraction algorithm, a new background subtraction method based on color similarity is presented The two-dimensional Otsu dynamic threshold algorithm is used in our new background subtraction algorithm.In order to obtain a better binary result, morphological method separate connected noise on the objects; inner outline filling method repair the holes inside the objects;Flood Fill algorithm is used to filter the image noise.Compared several camera’s calibration methods and based on the mirrors scene, a geometrical calculated camera’s parameters is presented. Through the consistency of the viewpoint, the contours sequence can be determined. These do important function for the 3D reconstruction.After reconstructed, all the information, such as the points position, the relation between each triangle, are saved in the Mesh data structure. The reconstructed 3D model can be rotated, transferred and zoomed in the OpenGL model browser. The model accuracy can be easily found in this browser.更多还原