【作者】 周晓明；

【导师】 马秋禾；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2011， 博士

【摘要】 机载激光雷达数据处理是Lidar应用系统的重要组成部分,大约有80%的工作量都集中在这上面,其中数据滤波DEM生产技术是最基础的数据处理工作,它对后续的数字产品生产、应用保障都起到关键作用,也是Lidar技术发展与研究的前沿性课题。本文围绕机载Lidar数据滤波技术,结合点云数据结构特征,进行了深入的研究和探索。本文的主要工作和创新点是:1.系统的阐述了机载Lidar的定位原理及其结构组成,介绍Lidar的主要数据结构及地物特征,分析数据处理的难点,为实际工程应用和后续的滤波算法设计提供理论依据。2.分析原始机载Lidar点云数据噪声点的形成原因,介绍新设备数据噪声形成的特点,针对传统高差阈值法需要大量的统计分析和实验判断的缺点,提出一种高程平面拟合与邻近点密度相结合的判断准则,采用迭代计算-自适应改变阈值参数的方式检测噪声点,改善了噪声剔除效果。3.改进了基于虚拟格网结构的形态学滤波方法。采用虚拟格网的数据组织方式,避免对原始数据进行内插格网损失精度。利用形态学梯度作为数据点是否进行开运算的阈值条件,提高对陡坎突变地形判断的准确性,对前后两次形态学开运算的结果进行二次判断,利用地物数据集区域的边缘特性作为地物识别特征,提高滤波的准确度,避免简单的阈值判断对地形数据的损失。实验分析证明,算法具有很强的适应性,在非城区的自然复杂地形数据具有良好的滤波效果。4.设计一种基于八叉树数据结构的点云聚类分割算法。利用八叉树的数据组织结构完成点云数据的聚类分割,对八叉树节点采用分割-融合两个步骤整合零散区域,改进区域分割的结果,提高准确性。同时在建立八叉树节点的过程中完成原始数据的索引组织结构,方便数据点集之间的查询计算,提高滤波效率。5.基于上述分割结果,提出一种结合区域拓扑结构信息与高差不连续性的滤波准则。对于分割结果的区域数据设计了一种拓扑结构信息,在区域块数据集类别判断上,考虑了区域之间高差不连续性及之间的拓扑结构关系双重判断标准,运用迭代计算的方式,逐渐剔除地物区域数据集以最终达到滤波的目的。该方法改善了传统聚类滤波中的地物判断缺点,在城区地形数据具有良好的表现。6.设计一种波形宽度阈值的全波形数据滤波方法。针对全波形数据的特点,用数学推导的方式证明其是高斯模型的波形叠加,采用最小二乘迭代优化的方式估计波形参数,利用波形宽度作为植被点和地形点的分类阈值,完成植被区域数据点的滤波。更多还原

【Abstract】 Airborne Lidar laser radar data processing is an important part of the application system, about 80% of the workload is concentrated in it, including data filtering DEM production technology is the most basic data processing work, its follow-up of digital products, Application security plays a key role, but also Lidar technology development and cutting-edge research topics. This paper focuses on the airborne Lidar data filtering and sorting technology, combined with point cloud data structure, in-depth study and exploration. The main work and innovations are:1. A systematic exposition of the principle of Airborne Lidar and the positioning of the composition, described Lidar data structure and surface features the main characteristics of data processing difficulties, the practical application and follow-up of the filtering algorithm to provide a theoretical basis.2. Analysis of Lidar point cloud data in the original noise causes the formation of rough almost to introduce the formation of new equipment, the noise characteristics of the data, the threshold elevation for the traditional method requires a lot of statistical analysis and experiments to determine the shortcomings, presents a fitting plane and elevation near the point of combining the density criterion, the iterative calculations - adaptive changes in the way of detection threshold parameter noise points.Experimental results show that the method can automatically detect outliers purpose of noise points, significantly lower risk of miscarriage of justice.3. Proposed structure based on virtual grid to improve morphological filter. The data using a virtual grid organization, to avoid interpolating the original data grid loss of accuracy. Morphological gradient as a data point is the opening operation of the threshold condition, improve the scarp terrain to determine the accuracy of the mutation, two consecutive morphological opening operation on the results of a second judge, using the edge of the regional characteristics of surface featuresObject Recognition as a feature, to improve the accuracy of filtering to avoid open operation blind misjudgment of the terrain data. Experimental analysis shows that the algorithm has strong adaptability to the natural complex in non-urban terrain with good filtering effect.4. Is proposed based on octree data structure of the point cloud cluster segmentation.Use of octree data structure to complete point cloud data clustering segmentation, using segmentation octree nodes - two-step integration of fragmented regional integration, improved segmentation results, to improve accuracy. At the same time in the establishment of octree nodes to complete the process of indexing the original data structure, between sets of data points to facilitate the mathematical calculations, to improve filtering efficiency.5. Segmentation based on the above results, we propose a new topology information and the regional height discontinuity filter. Segmentation results of the regional data for the design of a topology information, block data set types in the region to determine, considering the height difference between regions and between the discontinuity between the topology of the dual criteria, the use of iterative calculation method gradually removing the surface features of regional data sets in order to eventually achieve the purpose of filtering. This method solves the surface features of traditional clustering filter to determine shortcomings in urban terrain data with good performance, reliable and accurate filtering results.6. Presents a waveform width threshold of full-waveform data filtering methods. Full waveform data for the characteristics of a mathematical proof of its derivation is the superposition of Gaussian wave model using the least square iterative optimization method to estimate wave parameters, the use of waveform width points and terrain points as vegetation classification threshold, complete vegetation zone data point filtering.更多还原