【作者】 王力；

【导师】 华顺刚；

【作者基本信息】 大连理工大学 ， 机械设计及理论， 2006， 硕士

【摘要】 基于图像的建模技术(image-based modeling，IBM)是近年来兴起的基于图像的图形学方法的重要研究内容，也是计算机视觉的重要研究领域。该技术基于数码相机、数码摄像机等各种高性能数字化设备拍摄得到的数字图像，提取结构信息，重构场景或物体的三维几何模型。 未定标图像的三维模型重建技术是IBM中的一个重要分支，是指只需要场景或物体的不同视点序列图像就可以重构出场景或物体的三维几何模型。本文对该技术的研究包括立体匹配、相机自定标、特征点三维重建、纹理映射等内容。 立体匹配是计算机视觉研究的一项重要课题，本文研究了一种基于角点提取的立体匹配算法。首先使用最小同值分割吸收核(SUSAN)方法判断左右图像的边缘点是否为角点。然后根据角点间的特征相似度，建立起左右图像中被保留角点的匹配关系。利用加权归一算法估计基础矩阵的基础上，引入逐次去除异常匹配点，进行迭代计算，对基础矩阵求精。 相机自定标及对应特征点三维重建是三维重建技术中的关键问题。相机自定标包括相机内参数和相机外参数计算。本文根据基础矩阵估计相机焦距的方法，引入半定标矩阵来计算相机内参数；根据本质矩阵来计算相机的外参数。对应特征点的三维重建是根据三角测量的方法计算其投影矩阵，然后用奇异值分解求出特征点的三维齐次坐标。 本文中的纹理映射是将二维图像的纹理贴到三维框架表面，从而提高模型的可视性，达到“照片级”的视觉效果。本文以特征点为节点对二维图像进行三角剖分，然后将每个三角形纹理图像一一对应地贴到重建的三维模型表面。 基于上述研究，本文开发了一个基于未定标图像的三维模型重建系统，包括自动与互动方式相结合的图像角点提取及匹配，基础矩阵的估计及求精，相机的自定标，特征点三维重建和纹理映射等模块，能够得到较好的三维模型重建结果。更多还原

【Abstract】 The technique of image-based modeling is important content in computer graphics and computer vision. This technique can extract information of construction, reconstruct 3D model from digital images which captured by DC, DV and some other digital equipments.The technique for 3D surface reconstruction from two uncalibrated views is an important branch in image-based modeling, which only needs several images with different view-points. We study some techniques in this field, including stereo matching, camera self-calibration, feature points’ 3D modeling and texture mapping, etc.Stereo matching is an important task for the study of computer vision. A stereo matching algorithm based on corner detection is studied. First, the smallest univalue segment assimilating nucleus(SUSAN) approach is used to detect whether the pixels on the edges in the left and right images are corners, and the correspondence relationship by the retained corners is established between the corresponding corners of the left and right images according to the the similarity of the features or interactively. Then the fundament matrix is estimated from matched corners by an improved weighted linear algorithm, which is based on the epipolar geometry and the absolute conic theory.The camera self-calibration and the featrue points’ 3D modeling are two key techniques in image-based modeling. Camera self-calibration includes the computation of camera intrinsic parameters and camera external parameters. Based on fundament matrix and semi-calibrated matrix, the camera intrinsic parameters are computed. Then the external parameters are coumputed by essential matrix. Feature points’ 3D coordinates are computed through singular value decomposition of projector matrix, then compute projector matrix by triangulation.A realistic 3D surface model is built by mapping the triangular textures to the 3D structure surface. Triangular textures are acquired by dividing 2D image with Triangulation.We have developed a 3D surface reconstruction system from uncalibrated views, which is based on above various kinds of improved algorithms. This system is proven to be effective and satisfactory.更多还原