基于标记点过程的机载激光扫描点云建筑物提取

发布时间：2018-04-01 03:02

本文选题：机载激光扫描　切入点：建筑物提取　出处：《武汉大学》2013年博士论文

【摘要】：经过近二十多年的发展,机载激光扫描系统(Airborne laser scanning system, ALS)已经从最初的实验室研究阶段发展成为成熟的商业产品,具备数据采集速度快、处理周期短、高精度、高密度、获取成本较低等优点,能够直接、快速的采集大面积区域的空间三维信息,正日益成为空间数据采集技术的一个新的发展方向。机载激光扫描数据被广泛的应用于3D城市建模、电力线走廊三维制图、数字高程模型生成、植被检测以及环境研究等领域。建筑物作为人们生活和工作的重要场所,是城市空间中的重要实体,其位置边界信息是地籍图生成、建筑物三维重建、地图更新以及变形监测等应用方面的宝贵数据源。因此,研究机载激光扫描数据中的建筑物目标提取具有非常重要的意义。从机载激光扫描数据中提取建筑物目标一直是摄影测量与遥感、计算机视觉等领域的研究热点。近年来,国内外对基于机载激光扫描数据的建筑物目标提取进行了广泛的研究。但是由于建筑物及其周围地形的多样性和复杂性,使建筑物的提取具有较多的困难,至今还没有一种适用于任何数据质量以及任何场景的建筑物提取方法。同时由于机载激光扫描数据具有离散随机性以及分布不均匀性等特性,对建筑物的提取也带来了一定的困难。针对上述问题,本文研究了利用标记点过程从机载激光扫描数据中直接提取建筑物的方法。该方法是一种基于目标的、稳健的且能准确从点云数据中提取建筑物目标的方法。该方法首先构建出面向建筑物提取的Gibbs能量模型,然后对该能量模型进行优化求解以获取初步的建筑物区域,最后对初步获取的建筑物区域进行精细处理,从而得到精确的建筑物轮廓。本文的主要研究内容如下： 1、介绍了本文的研究背景和意义以及机载激光扫描系统及其应用。针对机载激光扫描数据的特点以及目前建筑物提取的难点,提出了本文的研究目标。针对机载激光扫描数据中建筑物目标的提取以及标记点过程方法用于几何目标提取这两个方面的研究现状进行了综述,总结了目前从机载激光扫描数据中进行建筑物目标提取的难点和不足及其可能的发展趋势。 2、介绍了利用标记点过程方法进行几何目标提取的研究背景,并简要叙述了点过程和标记点过程的理论。根据建筑物在机载激光扫描数据中的几何形状特点,确定了将长方体作为建筑物的几何标记模型。根据建筑物在散乱点云中的结构特征以及建筑物目标之间的相互关系构建了能直接从点云数据中提取建筑物目标的Gibbs能量模型,有效的将建筑物目标的空间信息和空间关系引入到新构建的Gibbs能量模型中,为建筑物目标的准确提取奠定了基础。 3、针对构建出的Gibbs能量模型的函数属于非线性函数,难以从理论上进行分析并获取对应的解析解的情况,本文采用RJMCMC方法与模拟退火算法相结合的方法获取Gibbs能量模型的全局最优解。虽然该求解方法能够从任何的初始状态下收敛到全局最优解,但运行效率却比较低,故本文对该求解方法进行了优化,以有效的提高算法的运行效率,节省运行时间。 4、求解Gibbs能量模型后可以初步的获取建筑物区域。但是这些区域的边界可能并不精确,也不完整,同一个建筑物可能由多个区域构成,同时其中也可能存在一些错误提取的区域。因此本文对初步获取的建筑物区域进行精细化处理。利用建筑物目标的几何属性特征结合区域生长算法剔除错误提取的地面点、树冠点、噪声点以及树冠等非建筑物目标；然后对邻接区域进行合并并利用改进的凸包算法求取每个建筑物目标的精确轮廓。 5、采用ISPRS机构提供的基准测试数据验证本文提出的算法的准确性和有效性。对本文提出的算法中用到的参数进行分析和探讨,并对实验结果进行精度评价与分析。最后以F1Measure作为度量建筑物目标提取的精度评价标准,将本文提出的方法与其它的建筑物提取方法进行比较。详尽的实验结果及其评价数据证明了本文提出的算法的准确性和有效性。
[Abstract]:After nearly more than 20 years of development, the airborne laser scanning system (Airborne laser scanning system, ALS) from the initial stage of the development of laboratory research into a mature commercial product, with fast data sampling, short treatment period, high precision, high density, low cost to obtain advantages, direct, three-dimensional information acquisition in large area the regional fast, has become a new development direction of spatial data acquisition technology. Airborne laser scanning data is widely used in the 3D city power line corridor modeling, 3D drawing, digital elevation model generation, field detection of vegetation and environmental research. The building as an important place for people to live and work, is an important entity in the city space, the location of the boundary information is the cadastral map generation, three-dimensional reconstruction of the building, the number of valuable map updates and deformation monitoring applications Therefore, it is very important to study the extraction of building targets in the airborne laser scanning data.
The target of building extraction from airborne laser scanning data is photogrammetry and remote sensing, research hotspot in the field of computer vision. In recent years, domestic and foreign to the airborne laser scanning data of buildings based on target extraction has been extensively studied. But because the buildings and the surrounding terrain for the diversity and complexity of the building extraction has more the difficulties, there is not an applicable to any data quality as well as building any scene extraction methods. At the same time as the airborne laser scanning data has the characteristics such as uneven distribution of discrete random and extraction of buildings, also brought some difficulties. To solve the above problems, this paper studies the marked point process method of direct extraction of buildings from airborne laser scanning data. This method is based on the target, robust and accurate from the point cloud data Method for extracting building targets. Firstly, construct the Gibbs energy model to building extraction, then the energy model is optimized to obtain the initial building area, and finally obtain preliminary fine processing on building area, to obtain accurate outline of buildings. The main contents of this paper are as follows:
1, introduces the research background and the significance as well as the airborne laser scanning system and its application. According to the characteristics of airborne laser scanning data and the current difficulties of building extraction, put forward the research goal of this paper. Point extraction and marking for the building by airborne laser scanning data process method for target geometric object extraction research status of the two the paper summarized, summed up the current from airborne laser scanning data were extracted and difficult building problems and the possible development trend.
2, introduces the research background of geometric object extraction using marked point process, and a brief description of the process and the marked point process theory. According to the geometry characteristics of the building in the airborne laser scanning data, determine the cuboid as geometric marking model. According to the building the relationship between structural features in the scattered the point cloud and the building constructed Gibbs model can extract building energy directly from the point cloud data, the spatial information and spatial relationship building is introduced into the Gibbs model in the construction of the new energy, laid the foundation for building the accurate extraction of target.
3, according to the function of the Gibbs energy model constructed belongs to nonlinear function, it is difficult to analyze from the theoretical analysis and obtain the corresponding solution, the global optimal method using RJMCMC method and simulated annealing algorithm for Gibbs energy model solution. Although this method can solve the convergence from the initial state to any the global optimal solution, but the efficiency is relatively low, so this paper optimizes the method, in order to effectively improve the efficiency of the algorithm, save the operation time.
4, to solve the Gibbs energy model can obtain building preliminary. But these boundaries may not be accurate, is not complete, the same building may consist of multiple regions, which may have some error extraction area. So this paper gets preliminary refinement of geometric properties of building area. The building features by combining region growing algorithm to eliminate the errors from the ground point, crown point, noise and other non target crown buildings; then the adjacent regions are merged and obtain the precise outline of each building by using improved convex hull algorithm.
5, the accuracy of benchmark data validation using ISPRS offers and effectiveness of the algorithm proposed. To analyze and discuss the parameters used in the proposed algorithm, and the accuracy evaluation and analysis of the experimental results. Finally, using F1Measure as the measure of building the target extraction accuracy evaluation standards and methods will be presented in this paper and other building extraction methods were compared. The experimental results detailed proved the accuracy of the proposed algorithms and the effectiveness and evaluation of the data.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TU19;P225.2

【参考文献】