【作者】 徐伟；

【导师】 刘晓平；

【作者基本信息】 合肥工业大学 ， 计算机软件与理论， 2009， 硕士

【摘要】 三维重建技术是计算机视觉、人工智能、虚拟现实等前沿领域的研究热点和难点,是人类在基础研究和应用研究中面临的重大挑战之一。这些需求导致了基于图像建模技术的产生,作为一种新兴的建模方法,此项技术受到了广泛的关注,它结合了计算机图形学、计算机视觉、摄影测量学、射影几何等许多学科的理论和成果。其最大的优点是可以直接从图像中重建几何模型,并抽取纹理和光照信息,从而简单、快速地构建出具有“照片级”真实感的三维模型。本文的研究属于基于图像建模技术的范畴,主要讨论如何将基于几何特征的摄像机自标定和基于多幅图像的三维重建方法应用于基于图像建模技术中。文章主要具体研究以下三点:1)基于图像灭点的摄像机自标定,首先推导得出图像上主方向灭点的像素位置与摄像机矩阵之间的约束关系,并根据摄像机视图中灭点空间几何约束,对摄像机矩阵进行了进一步的约束,进而直接求出在相差一个度量变换或一个欧式变换下的摄像机矩阵。2)基于单个空间平行四边形的摄像机标定的方法。该方法利用空间平行四边形各个顶点在摄像机坐标系下的相对深度比,得到各点在摄像机坐标系下坐标的解析表达式,最后利用四边形几何性质,列方程求解出摄像机参数和各点的相对深度,从而完成了标定。3)多幅图像基于灭点性质三维建模的方法。该方法利用灭点的属性推导出了摄像机中心在世界坐标系下的坐标位置的解析表达式,再利用两幅图像之间的空间几何约束关系,从空间约束关系中反算出物体各点的空间位置从而重建出物体几何模型,最后进行纹理映射,完成重建。更多还原

【Abstract】 3D reconstruction becomes the hot and difficult research direction in Computer Vision, Artificial Intelligence, Virtual Reality and other visualization fields. It is also one of the major challenges in basic and application research. With the rapid development in 3D reconstruction, Image-Based Modeling as a novel modeling method is coming out. Image-Based Modeling which combines many results in Computer Graphics, Computer Vision, Photogrammetry, Projective geometry has drawn increasing attentions in recent decades. The biggest merit of it is that posture and texture information can be directly extracted from images to reconstruct geometric models with photo-reality simply and rapidly. The research of this dissertation belongs to the field of Image-Based Modeling, and aims at applying camera self-calibration and object reconstruction based on multi-images to IBM. The main contents are as follows:1) In using vanishing point for calibration aspect, firstly, the dissertation shows constraint relationships between the pixel coordinates of vanishing points and camera matrix. Secondly, it constrains camera matrix which can be enhanced according to geometric restriction of vanishing points. Finally, camera matrix which lacks a Metric or Euclid transform can acquired and then decompose this matrix to determine intrinsic and extrinsic parameters.2) In using parallelogram for calibration aspect, every point’s relative depth rate is calculated in terms of the parallelogram geometry relationship. Meanwhile, every point’s analytical expression in camera coordinates can be acquired. Based on the analytical expression the focal length and other camera parameters can be drawn in terms of series functions.3) In object 3D reconstruction aspect, this dissertation shows a brand new method which reconstruct models from multi-images. At first, it use the properties of vanishing points for calculating position of camera center in world coordinates. In addition, model posture and position can be acquired by using geometry restriction of the camera. At last, the texture from images is extracted and rectified.更多还原