本文瑞典乌梅大学（作者：MonaForsman）的硕士论文，共71页。

有多种自动方法可用于创建从二维照片中提取的三维点云。在许多情况下，得到的结果是一个稀疏的点云，场景中的分布不均匀。确定一个物体在两幅图像中同一点的坐标后，利用摄像机数据和相对方位的知识，可以计算出该点的三维位置。由分布不均匀点云创建的模型可能会在稀疏区域丢失细节和精度。

本文的目的是研究点云的致密化方法。本文对不同的匹配点对提取方法进行了文献研究，并利用一套改进技术实现了最小二乘模板匹配（LSTM），在一组具有不同难度的不同场景上对实现方法进行评估。LSTM是通过对图像中密集的点网格进行处理来实现的，Wallis滤波用于增强对比度。利用优化后的参数对匹配点对应关系进行评价，以保持良好的匹配关系，并剔除不良的匹配关系，其目的是在图像中或类似平面的表面上查找靠近平面的细节。为了提高匹配点的质量，对LSTM进行了一组扩展。对同一模板，通过变换归一化互相关（TNCC）和多种子点（MSP）方法对种子点进行改进，然后测试它们是否收敛到相同的结果。Wallis滤波用于提高图像的对比度，通过与其它优化参数的相关性和x、y方向标准差的比较，评价了提取点的质量。如果某个点被否定，则存在使用较大模板尺寸重新尝试的选项，称为自适应模板尺寸（ATS）。

Several automatic methods exist forcreating 3D point clouds extracted from 2D photos. In many cases, the result isa sparse point cloud, unevenly distributed over the scene. After determiningthe coordinates of the same point in two images of an object, the 3D positionof that point can be calculated using knowledge of camera data and relativeorientation. A model created from a unevenly distributed point clouds may lossdetail and precision in the sparse areas. The aim of this thesis is to studymethods for densification of point clouds. This thesis contains a literaturestudy over different methods for extracting matched point pairs, and animplementation of Least Square Template Matching (LSTM) with a set ofimprovement techniques. The implementation is evaluated on a set of differentscenes of various difficulty. LSTM is implemented by working on a dense grid ofpoints in an image and Wallis filtering is used to enhance contrast. Thematched point correspondences are evaluated with parameters from theoptimization in order to keep good matches and discard bad ones. The purpose isto find details close to a plane in the images, or on plane-like surfaces. Aset of extensions to LSTM is implemented in the aim of improving the quality ofthe matched points. The seed points are improved by Transformed NormalizedCross Correlation (TNCC) and Multiple Seed Points (MSP) for the same template,and then tested to see if they converge to the same result. Wallis filtering isused to increase the contrast in the image. The quality of the extracted pointsare evaluated with respect to correlation with other optimization parametersand comparison of standard deviation in x- and y- direction. If a point isrejected, the option to try again with a larger template size exists, calledAdaptive Template Size (ATS).

1 引言

2 相关理论知识

3 致密化方法概述

4 具体实现

5 实验设计

6 实验结果

7 讨论

8 结论

9 未来工作展望

10 致谢

更多精彩文章请关注公众号：