KMS XINJIANG INSTITUTE OF ECOLOGY AND GEOGRAPHY，CAS
| 三维航空影像目标地物对象化分类研究 / 段永超著 ; 杨辽指导 | |
| 段永超 | |
| Subtype | 硕士 |
| Thesis Advisor | 杨辽 |
| 2016 | |
| Degree Grantor | 中国科学院大学 |
| Place of Conferral | 北京 |
| Degree Discipline | 测绘工程 |
| Keyword | 点云过滤 建筑物提取 道路提取 面向对象 |
| Abstract | 随着数字摄影测量技术的发展，利用有人或无人飞机挂载相机，进行高分辨率航空影像数据的获取变得也来越普遍。如何从这些含有丰富细节信息的高分辨率航空影像中对目标地物进行识别，提取，分类等操作，已经成为在摄影测量与遥感、计算机视觉等诸多领域研究的重点和热点,具有十分重要的理论和现实意义。建筑物作为城市的重要组成部分，对其进行正确识别和提取，在城市规划和建设中具有十分重要的意义。现阶段基于高分辨率航空影像的建筑物提取的研究，大部分是基于二维遥感影像数据，进行手动或半自动提取。这样的操作容易产生工作量大，效率低且不能充分利用航空影像中蕴含的三维信息等问题。道路作为衡量城市发展的主要因素之一,也是城市发展过程中更新最快的要素之一，道路的发展会直接影响城市功能用地的布局。本文从含有丰富细节信息的高分辨率航空影像中提取道路信息，以弥补传统提取识别方法中的一些不足，比如极易受到周围环境例如植被、车辆、阴影的影响，容易造成漏分错分，边缘模糊等。 针对以上这些问题，论文利用多种方法充分挖掘影像中的三维细节信息，借助航空影像数据的真数字正射影像（True Digital Orthophoto Map, TDOM）、数字高程模型（Digital Elevation Model, DEM）、数字表面模型（Digital Surface Model, DSM）等产品以及关键地物的光谱，形状、纹理等特征信息，进行点云过滤，从而快速高效，准确地进行关键地物的识别提取。主要完成了以下内容： （1） 对关键地物进行点云过滤。高分辨率航空照片生成的DSM在丰富影像细节信息的同时也放大了噪声。论文充分利用DSM的高程信息，以及相关光谱，纹理，形状等信息对关键地物提取时产生影响较大的噪声进行过滤，得到比较精细的DSM模型。 （2） 建筑物的自动识别提取。在进行建筑物识别提取时，植被对建筑物提取影响在众多因素中影响是最大，因此考虑先过滤掉植被再进行建筑物的提取。针对航拍时中心投影所导致的建筑物倾斜以及相互遮挡现象，论文采用通过重叠的航空影像提取TDOM作为底图，通过点云过滤，得到建筑物的DSM，以此作为高程限制来进行建筑物的提取。 （3） 道路的自动提取。在进行道路提取，在借助TDOM、DEM、DSM的同时，运用纹理、光谱、形状、背景上下文等特征对建筑物，植被等噪声的点云进行过滤。得到只含有少量噪声的道路地面DSM并以此作为辅助数据，结合面向对象技术进行城镇道路提取。 经过反复试验，论文中提出的技术和方法，均利用试验和比较等方式验证了其有效性和实用性。 |
| Other Abstract | With the development of digital photogrammetry technology, it becomes more and more common to get high resolution aerial image data through using manned or unmanned aircraft mounted camera. It has become a focus and hot in photogrammetry and remote sensing, computer vision, and many other research fields how to do the identification, extraction, classification and other operations of the target feature from these high-resolution aerial imagery containing rich detail information ,which has a very important theoretical and practical significance. Buildings, as an important part of cities, have a very important significance in the urban planning and construction. At the present stage, most of the researches about building extraction are based on high resolution aerial image is based on two-dimensional remote sensing data, manual or semi-automatic extraction. This operation easily leads to many issues, such as a large amount of work, low efficiency and can not fully use the three-dimensional information contained in aerial images. As one of the main factors to measure the development of the city, the road is one of the most updated elements in the process of urban development. The development of the road will directly affect the layout of urban functional land. In order to extract road information from high resolution aerial images with rich detail information and make up the traditional method that is easy to be influenced by the surrounding environment, such as vegetation, vehicle and shadow in extracting road, which easily leads to the wrong extraction, none-extraction and blurry edge. In order to solve above problems, the paper makes use of various methods to fully tap the three-dimensional image details that draws support from True Digital Orthophoto Map(TDOM), Digital Elevation Model(DEM), Digital Surface Model(DSM), and texture, spectrum, shape, background to filter out point cloud . In this way, we can accurately identify the key feature of extraction quickly and efficiently. The main novel points of this paper are described in the following: (1) Point cloud filtering for key features. The DSM generated by high resolution aerial images contains not only rich image detail information, but also the noise. The paper makes full use of the DSM height information, as well as the relevant spectrum, texture, shape and other information, so as to filter out the noise and get a more detailed DSM model. (2) Automatic recognition and extraction of buildings. The influence of vegetation on the extraction of buildings is the biggest, so we should filter out the vegetation firstly. According to the result of the central projection aerial building tilt and mutual occlusion phenomenon, the paper uses TDOM extracted by overlapping aerial images as a base map. Then we can get the DSM of buildings through point cloud filtering, as a height restriction to the extraction of buildings. (3) Automatic extraction of roads. This paper draws support from TDOM, DEM, DSM, and texture, spectrum, shape, background to filter out point cloud of building and vegetation. When getting the road DSM of little noise as height limit, we can make use of object-oriented technology for urban road extraction. After repeated experiments, the efficiency and robustness of the proposed technology and the methods are evaluated via comparative experiments and real practical applications. |
| Subject Area | 测绘工程 |
| Language | 中文 |
| Document Type | 学位论文 |
| Identifier | http://ir.xjlas.org/handle/365004/14728 |
| Collection | 研究系统_荒漠环境研究室 |
| Affiliation | 中科院新疆生态与地理研究所 |
| Recommended Citation GB/T 7714 | 段永超. 三维航空影像目标地物对象化分类研究 / 段永超著 ; 杨辽指导[D]. 北京. 中国科学院大学,2016. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment