基于影像匹配技术的点云数据精简算法研究

发布时间：2018-10-04 21:47

【摘要】：伴随着“智慧城市”地提出,对三维建模技术提出更高的要求是推进“智慧城市”发展的必要前提。虽然激光三维扫描技术已经存在且广泛应用,但是在获取大范围的城市区域数据及建立实景三维场景方面存在很多不便。在无人机平台的快速更新下,倾斜摄影技术得到迅猛发展。倾斜摄影技术具有获取数据高效、受环境影响小、成本低、得到模型真实及信息丰富等特点,已经被大范围地投入到土地管理与规划、农业、林业、考古等各个领域的应用中,对“智慧城市”的发展起着巨大地促进作用。基于倾斜摄影测量系统中得到三维点云数据的关键技术为多视影像匹配技术,三维点云数据带有丰富的表面纹理信息,存在较强的、直观的可视性,对于三维模型表面重建效率的提高与成本的降低有很大的帮助。然而影像匹配点云数据量庞大,对于后期数据处理、管理等极为不便,因此如何实现影像匹配点云数据地精简,以最少的点可以较好地表示三维模型就是论文主要研究内容。论文首先介绍了倾斜摄影技术的基本原理以及获取影像匹配点云数据的基本流程,然后通过Photoscan软件得到初始点云数据,对其进行预处理得到要进行精简研究的实验数据。主要研究内容为:对于预处理后得到的单栋建筑点云实验数据,选择了三种经典点云精简算法,包括随机采样法、曲率采样法、均匀格网法,分别对比了三种方法在不同精简率下的点云显示效果,分析了该三种方法存在的缺点与不足,在此基础上提出了两种算法组合的改进方法,其基本思路为:首先构建三角格网,然后求出三角格网法向量,并根据三角格网法向量求出每个点的法向量,求出相邻点法向量夹角,根据法向量夹角确定一临界值对点云数据进行精简,对点间法向量夹角比该临界值大的点数据保留不动,对夹角小于该临界值的点数据采取等距离采样的方法,通过设定不同的采样间距达到要求的精简率。最后对改进后的算法在表面积、体积上进行评价,并进行3D偏差分析,证明改进后算法的点云精简效果要优于前三种传统算法。
[Abstract]:With the proposition of "intelligent city", it is a necessary prerequisite to push forward the development of "intelligent city" by putting forward higher requirements to 3D modeling technology. Although laser 3D scanning technology has been widely used, there are many inconveniences in obtaining large range of urban area data and establishing real scene 3D scene. With the rapid updating of UAV platform, tilt photography technology has been developed rapidly. Tilt photography has been widely used in land management and planning, agriculture, forestry, archaeology and other fields because of its high efficiency in obtaining data, low impact by environment, low cost, real model and abundant information. It plays an important role in promoting the development of "wise city". The key technology to get 3D point cloud data based on tilt photogrammetry system is multi-view image matching technology. 3D point cloud data has abundant surface texture information and has strong visual visibility. It is helpful to improve the efficiency of surface reconstruction and reduce the cost of 3D model. However, because of the huge amount of data of image matching point cloud, it is very inconvenient for the later data processing, so how to realize the simplification of image matching point cloud data, and how to represent the 3D model with the least points is the main research content of this paper. This paper first introduces the basic principle of tilt photography and the basic process of obtaining image matching point cloud data, then obtains the initial point cloud data by Photoscan software, and then preprocesses them to obtain the experimental data to be simplified. The main research contents are as follows: for the experimental data of single building point cloud, three classical point cloud reduction algorithms are selected, including random sampling method, curvature sampling method, uniform grid method. The results of point cloud display under different reduction rates are compared, and the shortcomings and shortcomings of the three methods are analyzed. On the basis of this, two improved algorithms are proposed. The basic ideas are as follows: firstly, the triangular grid is constructed. Then the normal vector of triangular grid is obtained, and the normal vector of each point is obtained according to the normal vector of triangular grid, the angle of the normal vector of adjacent points is obtained, and a critical value is determined according to the angle of the normal vector to simplify the data of point cloud. For the point data whose normal vector angle is larger than the critical value, the point data whose angle is smaller than the critical value is fixed. The method of equal distance sampling is used to obtain the required reduction rate by setting different sampling intervals. Finally, the improved algorithm is evaluated on surface area and volume, and 3D deviation analysis is carried out. It is proved that the improved algorithm is superior to the former three traditional algorithms in point cloud reduction effect.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P23

【参考文献】