机载LiDAR波形数据高斯分解及点云分类研究

发布时间：2018-06-13 07:31

本文选题：全波形数据 + 高斯分解　；参考：《武汉大学》2017年硕士论文

【摘要】：机载LiDAR系统作为一种主动遥感技术,可以直接量测激光扫描仪与地形之间的距离,获取地面点的三维坐标,在森林参数估计、三维城市建模、电力线检测和数字地面模型生成等方面有着广泛的应用。目前的机载LiDAR系统大多具有波形数字化能力,可以记录发射脉冲的整个反射回波信号。用户可以通过分析回波波形得到地物的额外特征信息,进一步研究波形参数在植被提取、地物分类等方面的应用具有很大意义。本文通过波形数据分解得到的波形参数特征,研究波形参数在点云滤波与地物分类方面的应用,探索更加自动、精度更高的地物信息提取方法,从而发挥波形数据的应用优势。本文的主要研究内容如下:(1)机载LiDAR波形数据分解。在研究总结了波形数据高斯分解方法的基础上,提出了一种波形横向高斯分解初始参数估计方法,该方法对于复杂的多次叠加波分解,能够得到较为精确的初始参数。在去除无效的高斯分量后,利用LM算法对初始参数进行优化,最后利用波峰位置作为地物点响应位置解算得到生产点云。实验表明,本方法能够有效地检测各种类型的回波信号,波形分解生成的点云数量更多,层次更加丰富。(2)波形特征辅助的点云滤波。由波形分解得到的不同地物点具有不同的波形特征,本文选择波形宽度信息作为辅助特征以提高点云滤波质量,首先利用宽度阈值去除明显的非地物点,然后利用宽度信息计算每个点的权重,利用多级加权曲面滤波方法进行点云滤波,实现高精度的地形重建。实验表明,该方法能够很好地去除低矮植被等非地面点的影响,一定程度上提高滤波的正确率。(3)融合波形特征和几何特征的点云分类。在波形分解的基础上,通过分析得到的参数信息(强度、宽度、回波次数),结合曲率、高程差几何信息,选择样本区域进行训练构建决策树,对滤波后的分解点云进行地物分类,将非地面点分为建筑物、树木和低矮植被三种类型。实验表明,对测试集分类的总体精度达到了 94.86%,各类地物都取得了较好的分类效果。
[Abstract]:As an active remote sensing technology, airborne LiDAR system can directly measure the distance between laser scanner and terrain, obtain 3D coordinates of ground points, estimate forest parameters, and model 3D cities. Power line detection and digital ground model generation are widely used. At present, most airborne LiDAR systems have the ability of waveform digitization, which can record the whole reflected echo signal of the transmitted pulse. By analyzing the echo waveform, the user can obtain the additional feature information of the ground objects, and further study the application of waveform parameters in vegetation extraction, classification of ground objects and so on. In this paper, we study the application of waveform parameters in point cloud filtering and ground object classification by decomposing waveform data, and explore a more automatic and accurate method for extracting ground object information, so as to give play to the advantages of waveform data application. The main contents of this paper are as follows: 1) decomposition of airborne LiDAR waveform data. On the basis of studying and summarizing the Gao Si decomposition method of waveform data, a method for estimating the initial parameters of waveform transverse Gao Si decomposition is proposed. This method can obtain more accurate initial parameters for complex multiple superposition wave decomposition. After the invalid Gao Si component is removed, the initial parameters are optimized by LM algorithm, and the production point cloud is calculated by using the peak position as the response position of the ground object. Experiments show that this method can effectively detect various types of echo signals, and the number of point clouds generated by waveform decomposition is more, and the level is more abundant. Different ground objects obtained from waveform decomposition have different waveform characteristics. In this paper, the waveform width information is selected as the auxiliary feature to improve the quality of point cloud filtering. Firstly, the width threshold is used to remove the obvious non-ground points. Then the weight of each point is calculated by using the width information and the point cloud filtering is carried out by using the multi-level weighted surface filtering method to realize the high-precision terrain reconstruction. Experiments show that this method can remove the influence of non-ground points such as low vegetation and improve the accuracy of filtering to a certain extent. On the basis of waveform decomposition, through analyzing the parameter information (intensity, width, echo frequency), combining the geometric information of curvature and elevation difference, we select the sample area to train to construct the decision tree. The decomposed point cloud after filtering is classified into three types: buildings, trees and low vegetation. The experimental results show that the overall accuracy of the classification of the test set is 94.866.All kinds of ground objects have achieved better classification results.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】