机载激光LiDAR点云数据滤波和分类算法研究

发布时间：2018-01-22 20:15

本文关键词： LiDAR 滤波分类二面角滤波决策树　出处：《首都师范大学》2014年硕士论文　论文类型：学位论文

【摘要】：激光雷达(LiDAR, Light Detection and Ranging)是一种向目标发射激光束,将接收的目标返回信号与发射信号比较,通过处理得到地物三维信息和地面空间特征信息的雷达系统[1-2]。近年来,因LiDAR技术可快速获取高空间分辨率地表三维信息、高自动化数据采集效率,从而广泛应用于地形测绘、城市建模等多个领域[3-5]。激光雷达数据是离散的三维点云,点云数据的应用明显滞后于激光雷达系统的硬件发展。如何快速地处理大量三维点云数据,获取关注的建筑物三维信息,实现建筑物的提取工作,是学者们研究的重点和难点。基于以上观点,本文深入了解总结了近十多年来激光LiDAR点云数据滤波和分类的相关方法。重点研究LiDAR点云数据的滤波和分类算法两部分,首先根据首、末次回波高程值,去除植被点,将剩下的点云数据规则格网化,提升点云的处理效率,随后,提出一种新的考虑到相邻三点间高程变化快慢程度的二面角滤波法,进行滤波,然后,通过滤波前后的高程值变化对实验区进行区域分割,引入地物二面角均值属性特征,并结合回波次数、高程值和回波强度三个参数作为判定地物分类的约束条件,构建决策树;最后,采用Alpha Shapes算法提取建筑物轮廓线,进行地表三维显示。本文的主要研究内容如下： 1.深入地对机载雷达系统的定位原理以及点云的数据结构进行了介绍；并且,总结了激光LiDAR点云数据的处理流程。根据首、尾次回波高程值,去除植被点,进一步将点云数据进行规则格网化,提升了LiDAR点云的处理效率。 2.在规则格网存储的LiDAR点云基础上,鉴于传统的计算两点间坡度值的滤波算法,在地形变化剧烈时难以确定坡度阈值的情况,本文提出一种新的考虑到相邻三点间高程变化快慢程度的二面角滤波法,首次将空间二面角的平面角余弦值,表达空间中相邻两平面相对位置的概念引入LiDAR点云数据滤波中。首先,基于地表的连续性提取LiDAR点云数据中的高程突变点,然后,分别统计高程突变点和非突变点的二面角余弦频数分布,采用交点处对应的余弦值和提取高程突变点迭代的最小坡度阈值来判定地面点、非地面点,最后引入数学形态学“开”算子,去除低矮植被,最终得到可靠的滤波结果。本文方法,针对复杂城区环境,在滤除大型建筑物的同时,能准确快速地获取地面点集。 3.本文通过滤波前后LiDAR点云数据的高程值变化进行区域分割：在点云数据滤波后,LiDAR数据点被分为地面点及非地面点,于非地面点集中采用区域增长法分割点云；采用二面角均值、回波次数、高程值和回波强度四个参数构建决策树,将实验区地物分类为建筑物、植被、地面、道路四个属性,在此基础上,采用Alpha Shapes算法提取建筑物轮廓线,实现地表的三维显示 4.本文选择了2块海地太子港的局部LiDAR点云数据作为试验区,在Visual studio2010中采用二面角滤波法,进行点云数据的滤波,并在现有滤波方法中,与“渐进三角网法”(TerraSolid-Scan软件)进行对比分析,验证了本文算法的可行性。统计了分类结果混淆矩阵及Kappa系数,对本文的分类精度进行评估。最后,在Visual studio2010中实现了实验数据建筑物的提取和三维显示。
[Abstract]:Laser radar ( LiDAR , Light Detection and Radar ) is a kind of radar system which emits laser beam to the target , compares the received target return signal with the transmission signal , and obtains the three - dimensional information of the ground object and the surface space feature information by processing . In recent years , the application of the LiDAR technology can quickly acquire the three - dimensional information of high spatial resolution surface and high automation data acquisition efficiency . The laser radar data is a discrete three - dimensional point cloud , and the application of point cloud data is obviously lagging behind the hardware development of the laser radar system . Based on the above viewpoint , this paper deeply understands the correlation method of laser LiDAR point cloud data filtering and classification for more than ten years . It focuses on the filtering and classification algorithms of LiDAR point cloud data . Firstly , according to the first and last echo elevation values , the vegetation points are removed , the remaining point cloud data rules are meshed and the processing efficiency of the point cloud is promoted . Then , a new method is put forward , which takes the three parameters of the echo frequency , the elevation value and the echo intensity to be used as the constraint conditions for judging the classification of the ground objects , and finally , the surface three - dimensional display is carried out by using the Alpha Shapes algorithm to extract the building outline . The main contents of this paper are as follows : 1 . The positioning principle of airborne radar system and the data structure of point cloud are introduced in - depth ; and the processing flow of laser LiDAR point cloud data is summarized . According to the first and tail echo elevation values , the vegetation points are removed , and the point cloud data is further regularly meshed , so that the processing efficiency of the LiDAR point cloud is improved . 2 . On the basis of the LiDAR point cloud stored in the regular grid , in view of the traditional filtering algorithm for calculating the slope value between two points , it is difficult to determine the slope threshold value when the terrain changes violently . 3 . This paper divides the elevation value of LiDAR point cloud data before and after filtering : After the point cloud data is filtered , the LiDAR data points are divided into the ground point and the non - ground point , and the point cloud is divided by the regional growth method in the non - ground point set ; 4 . In this paper , the local LiDAR point cloud data in Port - au - Prince of Haiti is selected as the test area . In Visual Studio2010 , the filtering of point cloud data is carried out , and compared with TerraSolid - Scan software in the existing filtering method , the feasibility of the algorithm is verified . The classification accuracy is evaluated by using the classification result confusion matrix and Kappa coefficient . Finally , the extraction and three - dimensional display of experimental data buildings are realized in Visual Studio2010 .

【学位授予单位】：首都师范大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【参考文献】