【作者】 张齐勇；

【导师】 岑敏仪；

【作者基本信息】 西南交通大学 ， 地图制图学与地理信息工程， 2009， 硕士

【摘要】 机载LiDAR(Light Detection and Ranging)系统是一种主动式对地观测系统,是70年代初开始发展的一门新兴技术。LiDAR代表了一种新的独立的技术,可低成本、高密集、快速度、高精度的直接获取三维地表数据,自动化生成数字表面模型,成为各种测量应用中深受欢迎的一个高新技术。在数字城市中,可视化是其中一项重要的研究内容,因此城市表面信息的提取显得十分重要。机载LiDAR系统虽可获取城市表面信息,但这些表面信息特征不一,包含的内容繁杂,对这些不加以分析处理难以实现准确的城市建模,因此,树木与建筑物必须在地形表面上各自表示出来。目前LiDAR点云数据的地物提取研究大多集中在建筑物提取上,而树木提取研究又大部分集中在森林地区,对城区树木信息提取研究非常稀少。树木作为城市表面信息的重要组成部分,在可视化中不可忽视。因此,从城市环境中利用LiDAR点云数据提取树木具有很大的应用价值。目前对于LiDAR点云的树木提取主要就是一个分割与分类的过程,绝大部分算法都是基于这种思想实现。分割大部分通过聚类的方式或者图像处理算法实现,分类可以采用监督或非监督分类实现。较为典型的如Secord的树木提取算法,也是基于分割与分类的思想实现LiDAR数据的分类。Secord首先融合遥感影像,通过计算LiDAR数据点与点之间的相似性,再利用区域增长算法分割LiDAR点云数据,最后计算每一分割区域的特征向量,再采用支持向量机(SVM)对分割结果进行分类。这些树木提取算法在分割过程中没有充分考虑LiDAR点云自身的特点,分割算法过于复杂。在分类过程中,很多需要选取样本进行训练,为此需要大量高质量的有标记样本,计算复杂度高。本文首先探讨了LiDAR点云数据处理效率的问题,通过对LiDAR点云数据建立空间索引,极大地提高数据处理效率。然后在分割与分类的基础上,分析LiDAR点云DSM表面特征,提出了基于区域增长与梯度分割的树木提取算法,直接对LiDAR点云首次回波进行处理,提取出树木脚点,最后利用对应的影像对树木提取结果进行评估,提取率为85.9%,正确率为86.8%。实验表明,算法能比较好地从LiDAR点云数据中提取出树木脚点。由于基于区域增长与梯度分割的树木提取算法会将建筑物顶面的异常点误判为树木点,这种错误比较明显,在此对改算法改进,提出了对LiDAR点云数据进行灰度形态学开运算,滤除掉大的建筑物屋顶区域的异常点,再利用区域分割与梯度分割的思想提取树木,改进后的算法得到的树木提取率为86.3%,提取正确率为87.9%。与基于区域增长与梯度分割的树木提取原算法相比,除树木提取率与正确率略有改善外,更重要的是算法能在城区树木提取比较准确的基础上,极大地避免了屋顶异常点被判为树木点的错误。更多还原

【Abstract】 Airborne Light Detection and Ranging (LiDAR) is an active Earth Observation System (EOS), its development goes back to the 1970s and 1980s, with an early NASA system and other attempts in USA and Canada. It represents a new and independent technology. The 3D surface data can be acquired in low-cost、high-density、rapid、high-precision, to generate Digital Surface Model (DSM) automatically. Now, LiDAR has become a new technique in the survey field.As an important research content of the development of digital city, the extraction of city surface information signifies much for visualization. Though the LiDAR system can acquire the city surface information, the information has made up of different objects. It is difficult to realize city modeling if the surface information had not been processed. Therefore, the trees or buildings have to be represented separately on the terrain surface. At present, the research about the surface information extraction from LiDAR cloud data mainly focus on the building extraction and the tree extraction relatively lacks. Tree as an important part of city surface information, its visualization cannot be neglected. Therefor, extracting the tree from urban area in LiDAR data has much application value.At present, most of the methods to extract trees from LiDAR data are the course of segmentation and classification. We can use clustering or the usual algorithms of digital image processing to segment the LiDAR data, following, the trees may be separated from the segmented data by supervised classification or unsupervised classification. These algorithms, such as Secord’s algorithm, have two-step methods for tree detection consisting of segmentation followed by classification. The segmentation is done by a simple region-growing algorithm with weighted features from aerial image and LiDAR. His algorithm’s judging standard is based on the similarity between the points. Then, a feature vector is defined for each segment. The classification is done by weighted support vector machines (SVM). These methods do not fully consider the characteristic of the LiDAR data during the segmentation, and rather complicated. During the classification, they need select many high quality marked samples, and the calculation complications is high.This paper discusses the efficiency of the LiDAR data processing firstly. The efficiency of data processing can be greatly raised by building the spatial index for LiDAR data. Then the DSM characteristics are analyzed base on the segmentation and classification, and a tree extracting algorithm is presented from the LiDAR data of the first return pulse in complicated urban environment. This algorithm based on region-growing and gradient threshold. Finally using the corresponding image to evaluate the experimental results, the tree extracted ratio is 85.9% and its accurate ratio is 86.8%. The experimental results show that new algorithm can extract trees accurately from LiDAR data in urban region. Analyzing the tree extraction results, there have some abnormal points on the roof that had been mistaken as tree points. This is the obvious mistake. So, this algorithm needs to improve, and gray open operation is implemented to remove the abnormal points on the building roof. Then the region-growing and gradient threshold value is used to extract trees. After improved, the tree extracted ratio is 86.3% and the accurate ratio is 87.9%. Compare with the original algorithm, besides there is a slight improvement in the tree extracted ratio and its accurate ratio, what is more important the improved algorithm can avoid the obvious mistake that the abnormal points are taken as tree points.更多还原