车载校园环境激光点云分类与移动目标识别

发布时间：2018-05-28 12:03

本文选题：激光扫描 + 数据解译　；参考：《武汉理工大学》2013年硕士论文

【摘要】：激光扫描测量技术是继GPS之后测绘领域诞生的一种高新测绘技术,能够快速高效地获取目标物体表面详细的三维空间信息,在数字城市、环境监测、交通仿真等领域为快速实现三维建模提供了一种全新的技术手段。因此,如何有效地处理激光扫描数据具有实际的应用意义。在此背景下,本文以校园真实环境作为实验场景,以车载激光扫描系统作为三维信息的获取方式,研究车载激光扫描数据的处理,重点解决激光点云的分类问题和移动目标的特征提取问题。围绕上述研究目标,本文主要做了以下三个方面的工作： 1.进行车载校园激光数据的实地采集和预处理选择了特定的校园场景,使用新型的HDL-64E S2三维激光扫描仪进行了实验数据的实地采集。车载校园激光数据的预处理包括原始激光数据的解译、激光点云的精简和激光点云的三维可视化三方面的工作。通过激光数据的预处理才能获得后处理阶段所需的“点云”。 2.进行车载激光点云的实际分类激光点云分类是利用一些可行的分类策略,将海量独立的空间点划分到一系列具有实际物理意义的类簇里,使离散的独立点具有实际的物理意义。在分类过程中,本文先对比分析了三种现有的分类策略的优劣,然后选择垂直高度作为主要分类因素、投影点密度作为次要分类因素,提出了一个改进的基于垂直高度的双阈值法进行激光点云的分类,该方法的核心是基于网格的统计分析。 3.进行移动目标的特征提取和目标个体的识别移动目标(行人和车辆)的特征提取和目标个体的识别是在点云分类的基础(结果集)上进行的。鉴于激光点的高度离散性和抽象性,很难直接从中提取地物的点、线、面特征,本文采用了间接的处理方式,提出了一种基于图像处理技术的移动目标特征提取及识别方法：利用水平网格进行投影,将三维激光点云降维转换得到二维二值灰度图像,然后运用图像处理技术进行相关的特征提取,再根据提取的特征进行目标个体的识别。运用本文所提出的点云分类方法和特征提取方法,较好地实现了车载校园激光点云的分类、移动目标的特征提取和目标个体的识别。
[Abstract]:Laser scanning measurement technology is a new technology of surveying and mapping , which is born in the field of surveying and mapping after GPS . It provides a brand - new technical means for quickly and efficiently acquiring the detailed three - dimensional space information of the target object surface . In this background , the paper studies the processing of vehicle - mounted laser scanning data in the field of digital city , environment monitoring , traffic simulation and so on , which focuses on the classification of laser spot clouds and the feature extraction of moving targets .

Based on the above research objectives , the thesis mainly focuses on the following three aspects :

1 . On - site collection and preprocessing of vehicle - mounted campus laser data

This paper selects a specific campus scene , and uses the new HDL - 64E S2 three - dimensional laser scanner to collect the experimental data . The preprocessing of the laser data of the vehicle - mounted campus includes the interpretation of the original laser data , the simplification of the laser point cloud and the three - dimensional visualization of the laser point cloud . The " point cloud " required for the post - processing stage can be obtained by the preprocessing of the laser data .

2 . Actual classification of vehicle - mounted laser point cloud

The classification of laser point cloud is to use some feasible classification strategies to divide the mass independent spatial points into a series of cluster with physical meaning . In the process of classification , this paper compares the advantages and disadvantages of three existing classification strategies , then selects the vertical height as the main classification factor and the projection point density as the secondary classification factor , and proposes an improved double threshold method based on the vertical height for the classification of the laser point cloud .

3 . Feature Extraction of Moving Objects and Recognition of Target Individuals

The feature extraction and recognition of moving objects ( pedestrians and vehicles ) are carried out on the basis of point cloud classification ( the result set ) . In view of the high degree of discreteness and abstraction of laser spots , it is difficult to directly extract the points , lines and surface features of the ground objects . In this paper , a moving target feature extraction and recognition method based on image processing technology is proposed : a horizontal grid is used for projection , the three - dimensional laser point cloud is reduced and converted to a two - dimensional binary grayscale image , then the relevant feature extraction is carried out by using the image processing technique , and the identification of the target individual is carried out according to the extracted features .

By using the point cloud classification method and feature extraction method proposed in this paper , the classification , feature extraction and target individual identification of the vehicle - mounted campus laser point cloud are well realized .
【学位授予单位】：武汉理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P225.2;TP391.41

【参考文献】