面向对象的车载激光点云建筑物立面识别与三维重建

发布时间：2019-03-24 12:54

【摘要】：本文以面向对象的点云分析理论为指导,开展车载激光雷达(MLS)点云数据处理和信息提取工作,主要围绕建筑物立面识别、建筑物立面重建两个主题开展相关研究。主要研究内容包括以下四点：(1)车载激光点云分割与合并本文采用法向量与距离约束进行点云分割,该方法的稳健性好,但是分割后存在过分割现象。为此,采用kd树来确定分割面片的临近关系,采用分割面片之间的法线向量角度差异与分割面片之间的距离约束作为合并准则,进行分割面片合并。(2)基于先验知识的建筑物立面识别与提取点云分类和目标识别方面通常使用的特征包括：尺寸、形状、位置、方向、色彩、拓扑关系等特征。鉴于建筑物立面通常垂直于水平面、面积较大、显著高于周边地物这三个特点,本文主要使用分割面片的法线向量与水平面夹角、面积、绝对高程和高程变化范围三个特征提取建筑物立面面片。(3)车载激光点云建筑物立面轮廊线提取由于车载激光扫描点云数据中存在大量的立面点,传统的基于不规则三角网的机载激光扫描数据轮廓线提取方法失去了效力。本文对该传统方法进行改进,以适用于车载激光扫描数据。(4)建筑物立面拓扑关系矫正由于常规的特征值法提取的建筑物立面信息不准确。为此,基于建筑物立面垂直于水平面、建筑物立面之间存在垂直或者平行关系等先验知识,根据建筑物立面的空间法线向量检测建筑物立面之间的拓扑关系,并对建筑物立面之间的拓扑关系进行校正。面向对象的MLS点云建筑物立面识别、建筑物立面重建具有优异的效果,稳健性高。并且可以适用于复杂场景区域的点云数据处理和分析,为MLS点云数据建筑物立面识别与重建提供了一种行之有效的方法。
[Abstract]:Guided by the theory of object-oriented point cloud analysis, this paper carries out the data processing and information extraction of (MLS) point cloud of vehicle-mounted lidar, mainly focusing on the two topics of building elevation identification and building facade reconstruction. The main research contents are as follows: (1) vehicle laser point cloud segmentation and merging this paper uses normal vector and distance constraint to segment point cloud. This method has good robustness, but there is an over-segmentation phenomenon after segmentation. Therefore, the kd tree is used to determine the proximity relationship of the segmented patches, and the distance constraint between the normal vector angles and the segmented patches is adopted as the merging criterion. (2) recognition and extraction of building elevation based on prior knowledge. (2) the features commonly used in point cloud classification and target recognition include size, shape, position, direction, color, topological relation and so on. Since the facade of a building is usually perpendicular to the horizontal plane and has a larger area, which is significantly higher than the surrounding features, this paper mainly uses the normal vector of the segmented surface and the angle and area between the horizontal plane and the normal vector. Three features, absolute elevation and range of elevation variation, are used to extract facade patches of buildings. (3) because there are a large number of vertical points in vehicle laser scanning point cloud data, there are a lot of vertical points in vehicle laser spot cloud elevation veranda extraction of building facade. The traditional method of airborne laser scanning data contour extraction based on irregular triangulation has lost its effectiveness. In this paper, the traditional method is improved to be suitable for vehicle-borne laser scanning data. (4) the topology relation of building elevation is corrected because of the inaccuracy of building elevation information extracted by the conventional eigenvalue method. Therefore, based on the prior knowledge that the building facade is perpendicular to the horizontal plane and there is a vertical or parallel relationship between the building facades, the topological relationship between the building facades is detected according to the spatial normal vector of the building facade. The topological relationship between the facades of buildings is corrected. Object-oriented MLS point cloud building facade recognition, building facade reconstruction has excellent effect, high robustness. And it can be applied to point cloud data processing and analysis in complex scene area, which provides an effective method for building elevation recognition and reconstruction based on MLS point cloud data.
【学位授予单位】：辽宁工程技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TU198

【参考文献】