多视点图像的三维重建系统的设计与实现
发布时间：2019-04-23 23:38
【摘要】：基于图像的三维重建技术是一门融合了计算机图形学、图像学、计算机视觉、模式识别等技术的一个逆向工程。该技术在社会生活生产中有着广阔的应用:包括机器人视觉、医学成像、虚拟展示、生活娱乐等领域。三维重建技术是从普通相机获得的二维图像形成原始场景三维构造的过程。在这个过程中首先通过特征点提取和匹配计算相机的位置和投影参数,然后通过对极几何的稠密点匹配获得三维场景的稠密点云,接着通过对稠密点云进行网格化处理,重建目标场景的三维多面体模型,最后将原始图像映射到多面体模型上获得逼真的三维纹理模型。实现基于图像的三维重建系统,需要掌握单孔照相机成像模型、照相机标定、对极几何等三维重建的基础理论。在特征提取和匹配方面使用了经典的尺度不变特征变化(SIFT)算法和归一化互相关的优化算法,并使用RANSAC算法对特征匹配进行优化处理,删除错误匹配点,提高匹配的准确性。在稠密点生成过程中使用了基于块的稠密算法(PMVS),该算法很好的处理纹理不完整的场景,使用Power Crust算法进行点云的三角化处理。在最后的纹理处理过程中,我们使用开源的多面体处理工具MeshLab将原始图像映射到三维模型表面来获得最终的实体模型。最后通过与激光扫描仪取得的结果进行比较,定量分析了本三维重建系统的结果。结果显示,本三维重建系统的精度达到了应用水平。
[Abstract]:Image-based 3D reconstruction is a reverse engineering which combines computer graphics, computer vision, pattern recognition and so on. This technology has a wide range of applications in the production of social life, including robot vision, medical imaging, virtual display, life entertainment and so on. Three-dimensional reconstruction technology is the process of forming three-dimensional structure of original scene from two-dimensional image obtained from ordinary camera. In this process, the camera's position and projection parameters are first calculated by feature point extraction and matching, then the dense point cloud of 3D scene is obtained by matching dense points of polar geometry, and then the dense point cloud is gridded. The 3D polyhedron model of the target scene is reconstructed. Finally, the original image is mapped to the polyhedral model to obtain a realistic 3D texture model. In order to realize the 3D reconstruction system based on image, it is necessary to master the basic theory of 3D reconstruction such as single hole camera imaging model, camera calibration, polar geometry and so on. In the aspect of feature extraction and matching, classical scale invariant feature change (SIFT) algorithm and normalized cross-correlation optimization algorithm are used, and RANSAC algorithm is used to optimize feature matching, to delete wrong matching points and to improve the accuracy of matching. In the process of dense point generation, the block-based dense algorithm (PMVS),) is used to deal with the scene with incomplete texture, and Power Crust algorithm is used to triangulate the point cloud. In the final texture processing process, we use the open source polyhedron processing tool MeshLab to map the original image to the surface of the 3D model to obtain the final solid model. Finally, the results of the three-dimensional reconstruction system are quantitatively analyzed by comparing with the results obtained by the laser scanner. The results show that the accuracy of the three-dimensional reconstruction system is up to the level of application.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2463908
[Abstract]:Image-based 3D reconstruction is a reverse engineering which combines computer graphics, computer vision, pattern recognition and so on. This technology has a wide range of applications in the production of social life, including robot vision, medical imaging, virtual display, life entertainment and so on. Three-dimensional reconstruction technology is the process of forming three-dimensional structure of original scene from two-dimensional image obtained from ordinary camera. In this process, the camera's position and projection parameters are first calculated by feature point extraction and matching, then the dense point cloud of 3D scene is obtained by matching dense points of polar geometry, and then the dense point cloud is gridded. The 3D polyhedron model of the target scene is reconstructed. Finally, the original image is mapped to the polyhedral model to obtain a realistic 3D texture model. In order to realize the 3D reconstruction system based on image, it is necessary to master the basic theory of 3D reconstruction such as single hole camera imaging model, camera calibration, polar geometry and so on. In the aspect of feature extraction and matching, classical scale invariant feature change (SIFT) algorithm and normalized cross-correlation optimization algorithm are used, and RANSAC algorithm is used to optimize feature matching, to delete wrong matching points and to improve the accuracy of matching. In the process of dense point generation, the block-based dense algorithm (PMVS),) is used to deal with the scene with incomplete texture, and Power Crust algorithm is used to triangulate the point cloud. In the final texture processing process, we use the open source polyhedron processing tool MeshLab to map the original image to the surface of the 3D model to obtain the final solid model. Finally, the results of the three-dimensional reconstruction system are quantitatively analyzed by comparing with the results obtained by the laser scanner. The results show that the accuracy of the three-dimensional reconstruction system is up to the level of application.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前7条
1 陈树;王磊;;一种改进的基于RANSAC方法的SIFT特征匹配[J];信息技术;2016年12期
2 张艳;李强;;基于逐点插入法生成Voronoi图的算法研究及实现[J];黑龙江工程学院学报;2016年05期
3 谢利民;黄心汉;宋展;;基于不同光照条件的三维重建算法[J];华中科技大学学报(自然科学版);2013年S1期
4 Nguyen Tien Thanh;刘修国;王红平;于明旭;周文浩;;基于激光扫描技术的三维模型重建[J];激光与光电子学进展;2011年08期
5 刘雪娜;;三维点集Voronoi图的算法实现[J];计算机辅助工程;2006年01期
6 孟晓桥,胡占义;摄像机自标定方法的研究与进展[J];自动化学报;2003年01期
7 丁永祥，夏巨谌，，王英，肖景容;任意多边形的Delaunay三角剖分[J];计算机学报;1994年04期
相关博士学位论文 前2条
1 张学贺;基于双目视觉的六足机器人环境地图构建及运动规划研究[D];哈尔滨工业大学;2016年
2 康来;基于图像点特征的三维重建方法研究[D];国防科学技术大学;2013年
相关硕士学位论文 前3条
1 张也;基于局部降维的带约束三维Delaunay三角网构建算法研究[D];东南大学;2016年
2 戴嘉境;基于多幅图像的三维重建理论及算法研究[D];上海交通大学;2012年
3 李竞超;基于立体视觉的三维重建[D];北京交通大学;2010年
本文编号：2463908
本文链接：https://www.wllwen.com/shoufeilunwen/xixikjs/2463908.html
