基于旋转平台的多视角三维重建

发布时间：2018-04-28 01:10

本文选题：三维重建 + 多视角　；参考：《浙江大学》2013年硕士论文

【摘要】：近年来,随着影视动漫、逆向工程和工业检测等行业的迅速发展,三维重建作为其中核心技术之一,受到了普遍的关注和研究。本论文研究的三维重建基于一个旋转平台,该平台绕其中心轴做指定角度旋转,拍摄物体多个角度图像,获取物体完整的三维模型。论文围绕多视角三维数据获取和处理上的若干关键技术进行探讨,主要有三维数据的获取,不同视角三维数据的配准、融合等技术。三维数据的获取。本文设计了基于双目立体视觉的旋转平台系统,该系统由两个摄像机及一个可精确控制的旋转平台构成。针对双目立体视觉在物体无纹理区域,匹配难度较大的问题,结合时空立体视觉方法,用投影仪向物体表面投射条纹以增加纹理,实现了物体表面无纹理区域的精确重建。多视角三维数据的配准是构建完整三维模型的重要一步,目的就是寻找不同视角数据的空间位置转换关系,从而将这些空间点云数据统一到一个坐标系内。本文提出一种简捷有效的配准算法。算法根据标定板图像角点信息,推导出不同视角下点云数据的坐标转换关系,进而实现数据粗配准。目前使用最为广泛的精配准算法是两视角ICP算法,在此基础上,本文实现了多个视角的全局配准ICP算法,实现了各视角点云数据的高精度配准。多视角三维数据融合是三维重建的另一重要环节,指合并不同角度采集三维数据的重叠区域,目前被广泛应用的有基于网格缝合和基于隐式曲面的等融合方法。本文实现了基于泊松方程的隐式曲面融合算法,较好地重建了物体表面。实验结果表明,采用本文提出方法构建的三维模型精度高,真实感强且视觉效果令人满意。
[Abstract]:In recent years, with the rapid development of animation, reverse engineering and industrial detection, 3D reconstruction, as one of the core technologies, has received widespread attention and research. The 3D reconstruction in this paper is based on a rotating platform, which rotates around its center axis at specified angles, and takes multiple angle images of the object to obtain the complete 3D model of the object. In this paper, some key techniques of multi-view 3D data acquisition and processing are discussed, including 3D data acquisition, registration and fusion of 3D data with different visual angles. 3 D data acquisition. A rotating platform system based on binocular stereo vision is designed in this paper. The system is composed of two cameras and a precise controlled rotation platform. In order to solve the problem that binocular stereo vision is difficult to match in non-texture area of object, combining with spatio-temporal stereo vision method, the projector is used to project stripes to the surface of object to increase the texture, and the accurate reconstruction of texture-free area of object surface is realized. The registration of multi-view 3D data is an important step in constructing a complete 3D model. The purpose is to find out the spatial position transformation relation of the data of different visual angle, and to unify these spatial point cloud data into one coordinate system. This paper presents a simple and efficient registration algorithm. Based on the corner information of the calibration board image, the coordinate transformation relationship of point cloud data from different angles is deduced, and the rough registration of the data is realized. At present, the most widely used precision registration algorithm is the two-angle ICP algorithm. On this basis, the global registration ICP algorithm with multiple angles of view is implemented, and the high precision registration of the point cloud data of each angle is realized. Multi-angle 3D data fusion is another important part of 3D reconstruction. It combines overlapping regions of 3D data collected from different angles. At present, it is widely used in mesh suture and implicit surface based equal-fusion methods. The implicit surface fusion algorithm based on Poisson equation is implemented in this paper. The experimental results show that the 3D model constructed by the proposed method has high accuracy, strong reality and satisfactory visual effect.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【参考文献】