基于高程信息偏度平衡并顾及地形结构特征的机载LiDAR数据滤波方法的研究

发布时间：2018-04-17 11:41

本文选题：LiDAR + 点云滤波　；参考：《西南交通大学》2014年硕士论文

【摘要】：机载激光雷达系统是将激光测距(LDM)、全球定位系统(GPS)和惯性导航系统(INS)等三种技术集于一身的测量系统,它被用于获得地表的三维数据。利用直接采集获取的激光点云数据结合其对应的数码影像数据,经加工处理后便可以得到DEM、 DOM、DTM和DSM等多种丰富的数据产品。点云数据的滤波指的是从点云中分离出地面点集,这一过程是数据后处理中重要的组成部分,也是相关数据产品生产的基础。随着系统硬件的发展和人们对精度与准度的不断要求,滤波算法正面临着海量数据、多变的地形和复杂的地物等的挑战。在ISPRS于2003年对经典滤波算法测试分析之后,相继出现了不少新的滤波思路和方法改进,也有很多的研究和发展成果,但是阈值难以确定和对复杂多变的测区点云稳定性差仍然是目前大部分滤波算法存在的主要问题。因此,对真正能够准确、高效、适应性强的算法的研究是具有一定的价值和意义的。正是针对以上两点问题,本文通过研究总结现有滤波算法,改进Bartels等的偏度平衡算法,并利用ISPRS提供的参考数据样本进行实验分析。主要开展了以下工作： (1)研究了基于高程信息偏度平衡的滤波算法,归纳了国内外学者对该算法的相关改进,总结了各种改进后仍然存在的问题,并针对样本统计量的可靠性对点数的影响、平衡的终止条件难以满足、起伏地区迭代算法的适用性、缺少地形细节处理等四点不足,并针对这些不足提出了一种改进的算法,通过,对“起伏地区点云”将由高到低取点改为从低到高取点、将终止条件由sk≤0改为ske(某个很小的数),增强了算法对“起伏地区点云”处理的可靠性、收敛速度和适应性。 (2)总结了经典滤波算法的设计思路和对经典算法的相关改进。重点研究了基于TIN的逐渐加密算法,针对偏度平衡算法低矮地物难以滤除、地形结构特征被破坏的两点不足,进行了D-P算法地形特征点线提取、四叉树格网划分、地形特征点线约束下的种子点选取、添加点云边界辅助点、优先加密非特征格网和待定点排序等改进,增强了算法对待定点的判断准确性、对低矮地物的滤除效果和在处理“起伏地区点云”时保护地形特征的能力。 (3)选择了两块典型的点云数据,分别代表着平坦地区点云和起伏地区点云,利用改进后的算法进行滤波实验,将滤波结果与单独利用两种原方法的结果进行对比,从两类误差的角度评价了改进后算法的质量。实验表明,本文提出的滤波算法是可行的,它结合了偏度平衡与TIN加密的各自的优势。其中,平坦类点云通过偏度平衡算法从上往下划层后,使用TIN的逐渐加密算法对低矮地物的滤除有着较好的效果：而起伏类点云是从下往上依次划层,使每层数据是由低向高加密到TIN,并通过等高线分析提取地形特征点,对被加密的初始TIN进行约束,使得算法在较强阈值情况下既不会破坏地形结构也能滤除各个高程范围内的低矮地物。两类误差的分析表明,本文算法在在两类点云的滤波上均能很好地控制两类误差,并在控制第二类误差的前提下减少了第一类误差。
[Abstract]:Airborne laser radar system is the laser ranging (LDM), global positioning system (GPS) and inertial navigation system (INS) and other three kinds of technology in a measurement system, which is used for 3D surface data obtained. Laser point cloud data obtained by direct acquisition of its combination of digital image data should be. After processing can be DEM, DOM, DTM and DSM and other products. The rich data point cloud data filtering refers to the separation of the ground points set from point cloud data, this process is the important part in the postprocessing, is based on the data related to product production. With the development of hardware system and there have been calls on the accuracy and precision of the algorithm, is facing the challenge of massive data, changing the topography and complex features such as ISPRS in 2003. After the analysis of the classical filtering algorithm testing, there have been a lot of new filter Improvement ideas and methods, there are a lot of research and development results, but it is difficult to determine the threshold of survey area and point cloud stability of complex difference is still the main problem at present most filtering algorithm exists. Therefore, to really be able to accurately and efficiently, the research of adaptable algorithms is has a certain value and significance. It is for the above two problems, this paper summarizes the existing filtering algorithm, improved Bartels algorithm of skewness balancing of the sample and reference data provided by ISPRS are analyzed. The main work carried out the following:
(1) the elevation information filtering algorithm based on the balance of skewness, sums up the relevant domestic and foreign scholars on the improvement of the algorithm, summarizes the various improved problems and affects the reliability of the sample statistics on the number of the balance, the termination conditions are difficult to meet, the applicability of the ups and Downs Area iterative algorithm, the lack of terrain details processing four points, for these problems and proposes an improved algorithm, through to "ups and downs area point cloud" will be from high to low point changed from low to high points, the termination condition by SK = 0 to ske (a very small number). To enhance the algorithm on the "ups and downs area reliability point cloud processing, convergence speed and adaptability.
(2) summarizes the design ideas of classical filtering algorithm and improvement of the classical algorithm. Focus on the gradual encryption algorithm based on TIN, the skewness balancing algorithm is difficult to filter out low ground, two topography characteristics destroyed, the D-P algorithm for terrain feature points extraction, four fork tree grid. The seed point terrain feature points under the constraints of line selection, adding point cloud boundary auxiliary point, non priority encryption features of grid and fixed point ranking improved, enhanced the accuracy of the algorithm is to point to the low ground, filtering effect and protection of terrain features in the handling of the "ups and downs area point cloud" ability.
(3) chose two typical point cloud data, representing a flat area point cloud and downs area point cloud filtering experiment using the improved algorithm, comparing the filtering results and separately using two kinds of original method results from two kinds of error to evaluate the improved the quality of the algorithm.
Experimental results show that the algorithm proposed in this paper is feasible, it combines the skewness balancing and TIN encryption of their respective advantages. Among them, the flat point cloud by skewness balancing algorithm from the top row layer, using TIN encryption algorithm to filter out gradually low ground has a good effect: the ups and downs of point cloud from the bottom up in order to draw layer, each layer of data is from low to high encryption to TIN, and the extraction of terrain feature points through contour analysis, constraints on the initial TIN is encrypted, which makes the algorithm in strong threshold conditions without destroying the structure of terrain can also filter out each elevation within the scope of the low ground analysis. Two kinds of error shows that this algorithm can well control the two types of errors in two kinds of point cloud filtering, and reduces the error in the first premise to control second errors.

【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：P237;P225

【参考文献】