《【优秀硕士博士论文】点云降噪与采样技术的研究及应用.doc》由会员分享,可在线阅读,更多相关《【优秀硕士博士论文】点云降噪与采样技术的研究及应用.doc(73页珍藏版)》请在金锄头文库上搜索。
1、ABSTRACT硕士学位论文点云降噪与采样技术的研究及应用Reasearch and Application of Techniques on Point Cloud Smoothing and Sampling学科专业:机械电子工程2009年2月论文题目:点云降噪与采样技术的研究及应用学科专业:机械电子工程摘 要三维扫描技术和设备的发展,使得目前获取点云数据的速度和精度都得到极大的提高,可以在短时间内获得大量的点,有的甚至达到上千万个点。获取的点云数据中夹杂着很多噪声点和离群点,这些都不利于后期的曲面构建和点云数据与CAD模型比对,为此,本文对点云处理技术中的点云降噪、局外点检测和点云精简等
2、技术进行深入的研究,主要内容和成果如下:设计并实现了针对三维光学测量系统的点云处理方案,即首先去除点云数据中的噪声点,然后检测出局外点并剔除,最后将点云简化,得到数据量较小的高精度点云。实现了海量点云的去噪。研究了多种点云降噪算法,在此基础上,采用基于法向量的双边滤波算法。该算法先初步估算出点的法向量,然后利用KD-tree搜索点的邻域,根据点的邻域计算出点的偏移量,最后得出点的新坐标值。实现了点云数据中局外点的检测及剔除。采用了基于密度的局外点检测方法,该算法利用密度的概念,通过计算点的可达距离和点与邻域之间的距离,确定点的离群程度LOF(Local Outlier Factor),根据离群
3、程度判断该点是否为局外点。研究并实现了点云数据的精简。提出了一种新的可控精度的点云精简技术,该算法首先根据采样比率确定采样窗口的大小,然后沿着栅格线将整幅点云划分为一系列大小不同的采样窗口,最后根据给定的采样精度逐窗口精简高精度点云数据。经过试验、分析及工程应用,结果表明:在三维光学测量系统中较好实现点云降噪、局外点剔除和点云简化,效果良好。关 键 词:点云降噪;法向量;KD-tree;LOF;可控精度;点云精简论文类型:应用研究本研究得到国家“863”计划(2007AA04Z124)、江苏省科技支撑计划项目(BE2008058)资助。The research is funded by Nat
4、ional 863 Plan(2007AA04Z124)& Jiangsu Province Technology Supporting Planned Project(BE2008058).Title: Reasearch and Application of Techniques on Point Cloud Smoothing and SamplingSpeciality: Mechatronics Engineering Applicant: Yanhua PeiSupervisor: Assoc.Vice Prof.Jin LiangABSTRACTUntil now,the spe
5、ed and precision of the point cloud data acquisition have been greatly improved because of the development of 3D scanning technology and device,huge amount of points maybe reached super high can be obtained during the short time,and with many noises and outliers in the point cloud data,these are bad
6、 for the later surface construction and comparison of poind cloud data and CAD models.This paper has deeply researched the point cloud noise reducing,outlier detection and point cloud simplify.The main contents and achievements are listed as follows:The schem of point cloud processing has been desig
7、ned and implemented according to 3D optical measurement system, as is explained: the noises in point cloud data have been deleted firstly, then the outliers have been detected and rejected, the point cloud has been simplificated finally, and the small amount and high-accuracy point cloud has been ob
8、tained.The huge point cloud noise reducing has been realized. Many algorithms of noise reducing have been studied,on this basis,realized the huge data noise reducing by using bilateral filtering algorithm based on normal vector.First, the normal vector of points have been estimated primarily,and the
9、n KD-tree has been used for seaching the neighborhood of points, the deviation of points have been calculated by the neighborhood,finally the new coordinate of points have been obtained .The detection and rejection of outlier in point cloud data have been realized. Density-based outlier detection me
10、thod has been used. It utilizes the concept of density,the local outlier factor of point is determined by computing the reach-distance of point and the distance between point of the neighborhood,and judges if the point be a outlier by the local outlier factor of point.The simpification of Point clou
11、d data has been studied and implemented.A new and accuracy controllable technique of point data reduction is put forward,for this method,a sampling window is determined by the sampling rate firstly,the whole point cloud is divided into a series of windows with different sizes along the gridding line
12、s then,Finally the point data is reduced through the windows one by one basing on the given sampling accuracy.After testing, analyzing and the engineering application,the result indicated that: the point cloud noise reducing, outlier rejection and point cloud simplification can be realized well in t
13、he 3D optical measurement system,and has a good performance.KEY WORDS:Point cloud noise reducing;normal vetor;KD-tree;LOF;accuracy controllable;point cloud reductionTYPE OF THESIS: Application Research目 录绪论目 录1 绪论11.1 引言11.2 研究背景与国内外发展状况11.2.1 研究背景11.2.2 国内外发展状况21.2.3 国内外研究现状总结31.3 课题来源和研究意义31.3.1 课
14、题来源31.3.2 研究意义31.4 研究内容和技术路线41.4.1 研究内容41.4.2 技术路线41.4.3 可行性分析51.5 本文结构安排52 点云处理方案72.1 引言72.2 XJTUOM系统72.2.1 XJTUOM三维光学面扫描系统介绍72.2.2 点云处理所处的环节82.3 点云处理系统方案设计82.3.1 点云处理系统设计原则82.3.2 针对三维光学测量的点云处理系统方案92.3.3 点云处理目标的确定102.4 点云光顺技术研究102.4.1 三维几何去噪102.4.2 常见光顺去噪算法112.4.3 KD-Tree邻域搜索132.4.4 拉普拉斯光顺142.4.5 点
15、的法向量计算152.4.6 双边滤波算法162.5 小结183 局外点检测技术研究193.1 引言193.2 检测局外点的方法193.2.1 基于分布局外点检测193.2.2 基于深度局外点检测193.2.3 空间局外点检测203.2.4 基于聚类局外点检测203.3 基于距离的局外点检测203.4 基于密度的局外点检测223.4.1 PLY格式点云文件233.4.2 基于密度检测局外点253.5 小结284 点云采样技术研究294.1 引言294.1.1 点云精简算法分类294.1.2 点云精简算法评价304.2 常见点云精简算法304.2.1 包围盒法304.2.2 曲率采样法304.2.3 迭代法314.2.4 弦偏差法314.2.5 均匀网格法324.3 聚类法采样324.3.1 聚类法分类324.3.2 协方差分析法334.3.3 层次聚类法344.4 基于栅格线采样374.5 本章小结405 点云处理系统的试验分析及应用415.1 引言415.2 点云处理系统软件的实现415.2.1 软件的实现415.2.2 功能模块425.3 点云处理系统试验分析435.3.1