机载LiDAR点云数据的建筑物提取和模型规范化研究

发布时间：2018-03-22 13:38

本文选题：机载LiDAR　切入点：数据滤波　出处：《南京大学》2013年硕士论文　论文类型：学位论文

【摘要】：建筑物的识别与提取一直是测绘、遥感领域研究的热点。建筑物作为城区地面的重要特征在图像特征匹配、地形图更新、数字城市等领域起重要作用。机载LiDAR作为一种新兴的对地观测技术,能够快速实现地表三维信息的识别与提取。如何从海量的LiDAR点云数据中快速自动化地实现建筑物提取,具有十分重要的意义。本文在总结前人研究的基础上,提出了一种基于TIN模型的点云分割滤波算法以及一套基于轮廓线的建筑物边界多边形矢量化算法。主要研究内容和结论如下： (1)典型地物LiDAR点云空间分布规律：点云分布特征不仅与相应地物表面物质特性有关,还与其表面平坦与否相关。本文深入地分析了典型地物相应的LiDAR点云的空间分布规律,得出噪声点多呈孤立点分布、植被点云团簇状分布而方差较大、地面平坦而高程差较小、水体因吸收作用,数据点稀疏、地面突出物体积小,脚点数量少、建筑物点云分布规律,点密度均匀等结论。 (2)机载LiDAR滤波算法：在综合评价几类经典滤波算法的基础上,提出了一种基于TIN模型的点云分割算法。本算法通过构建TIN模型建立点云之间的空间邻接关系,采用聚类的思想对点云进行分割,使点云聚类成内部均匀的对象,然后根据对象的高程与方差过滤出建筑物点,完成滤波。选择了三个典型的研究区(10。左右的斜坡、含粗差的不规则复杂建筑物、居民区)进行实验,实验结果表明：斜坡的滤波总体精度达96.61%,Kappa系数达0.9154；复杂建筑物的滤波总体精度达91.69%,Kappa系数达0.8183；大片居民区的滤波精度达93.72%,Kappa系数达0.8076.说明本算法具有很好的抗粗差能力,对斜坡、复杂建筑物等也具有较好的滤波效果,适合城区建筑物LiDAR数据滤波。 (3)建筑物轮廓多边形矢量化：本文应用Alpha-Shapes算法追踪建筑物点集的轮廓线,实现了凹型建筑物及构造复杂建筑物的内外轮廓线的有效提取；其次,本文提出的改进的最小方向差模型拟合算法保证了所计算出的建筑物主方向与初始轮廓多边形所有线段之间的方向差达到最小。实验表明主方向估计值几乎无偏差的概率达69.4%,微小偏差的概率为26.5,大偏差的概率仅为4.1%,提取精度远远高于常用的基于Hough变换、主成分变换、统计直方图等方法。另外,建筑物轮廓多边形规范化方法对初步规范化后的多边形进行边界扩展,使扩展后的多边形线段能够定位于建筑物边缘点。
[Abstract]:The identification and extraction of buildings has been a hot topic in the field of surveying and remote sensing. As an important feature of urban ground, buildings are matched in image features and updated in topographic maps. As a new technology of Earth observation, airborne LiDAR can quickly realize the recognition and extraction of 3D surface information. How to quickly and automatically realize building extraction from massive LiDAR point cloud data. It is of great significance. On the basis of summarizing the previous studies, this paper proposes a point cloud segmentation filtering algorithm based on TIN model and a set of building boundary polygon vectorization algorithm based on contour. The main contents and conclusions are as follows:. 1) the spatial distribution law of LiDAR point cloud of typical ground objects: the distribution characteristics of point clouds are not only related to the surface properties of the corresponding ground objects, but also related to the flatness of the surface. The spatial distribution law of the corresponding LiDAR point clouds of typical ground objects is analyzed in depth in this paper. It is concluded that the noise points are mostly distributed in isolated points, the vegetation point clouds are clustered and the variance is large, the ground is flat and the elevation difference is small, the water body is absorbed, the data points are sparse, the surface protrusions are small in volume and the number of foot points is small. The rule of point cloud distribution and the uniform point density of the building are obtained. (2) Airborne LiDAR filtering algorithm: on the basis of synthetically evaluating several classical filtering algorithms, a point cloud segmentation algorithm based on TIN model is proposed. The spatial adjacency relationship between point clouds is established by constructing TIN model. The idea of clustering is used to segment the point cloud, so that the point cloud is clustered into an internal uniform object. Then, according to the height and variance of the object, the building points are filtered out to complete the filtering. Irregular and complex buildings, residential areas, with gross errors. The experimental results show that the overall filtering accuracy of slope is 96.61 and that of Kappa coefficient is 0.9154, that of complex building is 91.69 and that of Kappa is 0.8183, that of large residential area is 93.72and Kappa is 0.8076.The result shows that the algorithm has a good ability to resist gross error, and has good ability to resist gross error. Complex buildings also have better filtering effect, which is suitable for LiDAR data filtering of urban buildings. (3) Vectorization of building contour polygon: in this paper, Alpha-Shapes algorithm is used to trace the contour line of the building point set, which can effectively extract the inner and outer contour lines of concave building and complex building. Secondly, The improved least direction difference model fitting algorithm proposed in this paper ensures the minimum direction difference between the main direction of the building and all the lines of the initial contour polygon. The experimental results show that the estimated value of the principal direction is almost unbiased. The probability of difference is 69.4, the probability of small deviation is 26.5, the probability of large deviation is only 4.1, and the precision of extraction is much higher than that of Hough transform. In addition the method of building contour polygon normalization extends the boundary of the primary normalized polygon so that the expanded polygon line segment can locate at the edge of the building.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P225.2

【参考文献】