基于车载三维激光扫描数据的分类与建筑物提取

发布时间：2019-05-22 17:41

【摘要】：近年来城市三维空间信息的获取与应用发展日益成熟,已被广泛应用于城市建设与规划中。车载激光扫描系统可快速、自动、连续获取高精度的三维空间数据,作为一种新的数据获取手段被逐渐应用于地理信息产业中。车载扫描系统在获取数据时,可近距离获取城市街区中多种地物的三维空间信息。而在城市建设中,建筑物贯穿着整个城市,将车载激光扫描系统获取的建筑物点云数据进行快速的分割与提取显得尤为重要。本文对国内外研究现状进行了总结,介绍了车载扫描系统的构成和工作原理,介绍了点云数据的采集过程及点云数据的处理流程,总结了其他学者基于激光扫描数据所采用的分类方法,通过总结对比选取了本文方法。本文采取基于残差分析与区域生长的方法对点云数据分别进行了粗分类和细分类。首先本文在粗分类过程运用局部邻域法向量和基于平面拟合残差的平整度属性,将点云分割成不同区域如建筑物、地表、电杆、植被等。在进行细分类时利用建筑物的平面属性,将建筑物点云分类成不同的细节区域,如窗户、门、墙体等。在细节分类中进一步提取建筑的细节成分时,局部区域拟合残差用来决定一个点是否在一个平面区域内,法向量夹角来决定邻域点的相似程度。通过计算两个参数θ和St来限制区域生长过程,在生长过程中能通过St值来剔除噪声点,以达到分类效果。本文的分类方法在分类过程中不但能将建筑物中的细节部分进行提取,还能对噪声点有一定的识别能力,减少内存空间。对平面和非平面的点云数据有一定的提取效果。
[Abstract]:In recent years, the acquisition and application of urban three-dimensional spatial information has become more and more mature, and has been widely used in urban construction and planning. The vehicle laser scanning system can obtain high precision 3D spatial data quickly, automatically and continuously. As a new means of data acquisition, it has been gradually applied to the geographic information industry. When the vehicle scanning system acquires the data, it can obtain the three-dimensional spatial information of many kinds of ground objects in the city block at close range. In the urban construction, the building runs through the whole city, so it is particularly important to segment and extract the building point cloud data obtained by the vehicle laser scanning system quickly. This paper summarizes the research status at home and abroad, introduces the structure and working principle of the vehicle scanning system, and introduces the acquisition process of point cloud data and the processing flow of point cloud data. The classification methods adopted by other scholars based on laser scanning data are summarized, and the methods in this paper are selected by summary and comparison. In this paper, the rough classification and subdivision of point cloud data are carried out based on residual analysis and regional growth. Firstly, in the rough classification process, the local neighborhood normal vector and the smoothness attribute based on plane fitting residual are used to divide the point cloud into different areas such as buildings, surface, pole, vegetation and so on. The point clouds of buildings are classified into different detail areas, such as windows, doors, walls and so on, by using the plane properties of buildings. When the detail components of the building are further extracted from the detail classification, the local area fitting residual is used to determine whether a point is in a plane region or not, and the angle of the normal vector determines the similarity degree of the neighborhood points. By calculating two parameters theta and St to limit the growth process of the region, the noise points can be eliminated by St value in the process of growth, so as to achieve the classification effect. In the process of classification, the classification method in this paper can not only extract the details of the building, but also have a certain ability to recognize the noise points and reduce the memory space. It has a certain extraction effect on planar and non-planar point cloud data.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P225.2

【参考文献】