基于多源遥感图像与LiDAR数据的建筑物三维空间信息提取

发布时间：2019-03-08 15:37

【摘要】：随着经济、社会的飞速发展,城市规划及基础设施建设、环境监测及能源利用评估、军事侦察和目标精确打击等多种民事和军事应用对三维城市建模的性能要求越来越高。由于建筑物是三维城市模型中的最主要目标,利用遥感成像数据实现大范围建筑物三维建模所需信息(即建筑物三维空间信息)提取的相关研究一直是相关领域的研究热点。其中,遥感光学图像(卫星图像或航拍图像)是实现建筑物三维空间信息提取的最主要数据源。然而,受到关键结构检测及匹配等技术的制约,遥感光学图像建筑物三维空间信息提取的效率较低。近年来,随着机载激光雷达(Light Detection And Ranging,LiDAR)技术的不断成熟,机载LiDAR数据在建筑物三维空间信息提取方面体现出了与遥感光学图像的显著互补性,通过联合遥感光学图像和机载LiDAR数据能够显著增强建筑物三维空间信息提取的性能。然而,联合两种数据源进行建筑物三维空间信息提取的相关研究主要集中于联合的方式方法上,而对两种数据源在面向联合建筑物三维空间信息提取的优势能力挖掘方面的研究还存在很多亟待解决的问题。因此,本文以挖掘两种数据源的优势能力为手段,以联合实现建筑物三维空间信息的精确可靠提取为目的,分别从多源遥感光学立体像对三维空间位置信息提取、机载LiDAR数据建筑物三维空间结构信息复原、最优拓扑结构约束的建筑物三维空间信息联合提取等方面开展研究工作,并做出了如下几方面的主要贡献:(1)针对原始数据的信息表示能力问题,研究了提高遥感光学图像和机载LiDAR点云原始数据信息表示能力的方法。其中,在遥感光学图像方面,通过设计合理的地面控制点选择方法,提高了光学图像相机模型的三维空间位置信息表示精度;在机载LiDAR数据方面,通过详细分析机载LiDAR系统主要噪声源在数据采集过程中的误差传递,建立了机载LiDAR数据及对应栅格化图像的随机噪声表示模型,提高了机载LiDAR数据的三维空间结构信息表示能力。(2)针对传统光学图像三维空间信息提取过程中数据要求严格且精度受限的问题,提出了一种基于多源光学立体像对的高精度三维空间位置信息提取方法。首先,在广义最小二乘最优加权理论的指导下,通过详细的像素实际分辨率分析,构建了多源光学立体像对三维空间信息提取的数学模型。在数学模型求解阶段,由于该模型是一个观测值不平衡的不等精度广义线性方程组,在迭代求解过程中,提出了利用方差/协方差分量估计(Variance/covariance Components Estimation,VCE)技术实现精度控制的方法,有效提高了三维空间位置信息提取的精度。(3)针对建筑物屋顶机载LiDAR数据中结构性随机误差对建筑物三维空间结构信息提取精度和可靠性的影响问题,提出了一种基于结构化稀疏表示的建筑物屋顶机载LiDAR数据三维空间结构信息复原方法。在机载LiDAR数据栅格化图像结构性随机误差模型的指导下,通过充分考虑建筑物屋顶结构的稀疏性,提出了一种基于结构化稀疏表示的建筑物屋顶机载LiDAR数据三维空间结构信息复原框架,并给出了该框架的几种典型实现方法,有效实现了原始建筑物机载LiDAR数据三维空间结构信息的复原,为实现精确可靠的建筑物三维空间信息提取奠定了坚实的基础。(4)针对建筑物三维空间结构信息提取时,遥感光学图像受阴影及复杂环境细节信息影响严重,LiDAR数据提取阶跃特征/边缘定位不够精确且结构信息提取不够稳定的问题。提出了一种分级能量最小化的建筑物屋顶机载LiDAR数据全局最优平面分割方法,联合遥感光学图像的边缘信息可实现建筑物顶面拓扑结构的精确稳定提取,保证了最终提取三维空间信息的精确性和稳定性。
[Abstract]:With the rapid development of the economy, the rapid development of the society, the construction of urban planning and infrastructure, the assessment of environmental monitoring and energy utilization, the military reconnaissance and the accurate target of the target, the performance requirements of the three-dimensional urban modeling are getting higher and higher. Because the building is the main goal in the three-dimensional city model, the research on the information of the three-dimensional modeling of large-scale buildings (that is, the three-dimensional space information of the building) has been a hot topic in the relevant fields. The remote sensing optical image (satellite image or aerial image) is the most important data source for the information extraction of three-dimensional space of the building. However, the efficiency of the three-dimensional spatial information extraction of the remote sensing optical image building is low due to the technical constraints such as the detection and matching of the key structures. In recent years, with the development of the technology of the airborne lidar (LiDAR), the airborne LiDAR data has shown remarkable complementarity with the remote sensing optical image in the three-dimensional space information extraction of the building. Through the combined remote sensing optical image and on-board LiDAR data, the performance of the three-dimensional spatial information extraction of the building can be significantly enhanced. However, the research of three-dimensional spatial information extraction with two data sources is mainly focused on the joint mode method, and there are many problems to be solved in the research of the advantages of the two data sources in the three-dimensional space information extraction of the joint building. Therefore, in order to realize the accurate and reliable extraction of the three-dimensional space information of the building, the three-dimensional space structure information of the airborne LiDAR data building is recovered by combining the advantages of the two data sources as a means to realize the accurate and reliable extraction of the three-dimensional space information of the building, The research work is carried out in three-dimensional space information of the building with the optimal topological structure constraint, and the main contribution of the following aspects is made: (1) the ability problem is expressed for the information of the original data, The method of improving the information representation ability of the original data of the remote sensing optical image and the airborne LiDAR point cloud is studied. in that aspect of the remote sensing optical image, by designing a reasonable ground control point selection method, the three-dimensional space position information representation accuracy of the optical image camera model is improved; and in the aspect of airborne LiDAR data, Based on the detailed analysis of the error transfer of the main noise source of the airborne LiDAR system in the data acquisition process, the airborne LiDAR data and the random noise representation model corresponding to the rasterized image are established, and the three-dimensional space structure information representation capability of the on-board LiDAR data is improved. (2) A high-precision three-dimensional space position information extraction method based on multi-source optical stereo image pair is proposed for the problem of strict and limited data requirement in the traditional three-dimensional spatial information extraction process of the optical image. First, under the guidance of the generalized least-squares optimal weighting theory, the mathematical model of the three-dimensional spatial information extraction of the multi-source optical stereo image is constructed by the detailed analysis of the actual resolution of the pixels. In the solution phase of the mathematical model, because the model is a non-equal-precision generalized linear system with the unbalance of an observation value, the method for realizing the precision control by using the variance/ covariance component estimation (VCE) technology is put forward in the iterative solution process. And the accuracy of the three-dimensional space position information extraction is effectively improved. (3) In view of the influence of the structural random error on the information extraction precision and reliability of the three-dimensional space structure of the building, a method for restoring the structure information of the on-board LiDAR data in the building roof based on the structured sparse representation is presented. under the guidance of the on-board LiDAR data rasterized image structural random error model, the structure information recovery framework of the on-board LiDAR data three-dimensional space structure of a building roof based on a structured sparse representation is proposed by taking full consideration of the sparsity of the building roof structure, In addition, several typical implementation methods of the frame are given, and the restoration of three-dimensional space structure information of the original building on-board LiDAR data is effectively realized, and a solid foundation is laid for realizing the accurate and reliable three-dimensional spatial information extraction of the building. And (4) when the three-dimensional spatial structure information of a building is extracted, the remote sensing optical image is affected by shadow and complex environment detail information, and the LiDAR data extraction step characteristic/ edge positioning is not accurate and the structure information extraction is not stable enough. A global optimal plane segmentation method for the on-board LiDAR data of a building roof with minimum classification energy is proposed, and the edge information of the combined remote sensing optical image can be used for realizing the accurate and stable extraction of the topological structure of the top surface of the building, and the accuracy and the stability of the three-dimensional space information can be finally extracted.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】