多视影像密集匹配点云分类与测图应用研究

发布时间：2018-06-29 03:47

本文选题：倾斜摄影 + 点云分类　；参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：随着计算机视觉和摄影测量技术的发展,利用计算机视觉的运动恢复目标结构算法(Structure Form Motion,SFM)和多视图立体视觉重构算法(multi-view stereo,MVS),对多视图像进行匹配的倾斜摄影测量技术迅速的发展起来。多视倾斜摄影测量可以从多个角度对地面地物进行拍摄,能够获得对象除垂直方向外的侧面纹理信息,可以全面的反映对象的纹理属性。使用SFM和MVS算法对多视影像进行匹配,获得对象的稀疏点云、密集点云,并使用垂直影像制作正射影像。分别使用格网化数学形态学法和迭代三角网内插法对密集点云数据进行分类处理,分类后的地面点云生成DEM,并生成等高线,结合正射影像的地物数字化图,制作DLG。本文研究内容的方法和结果如下:(1)对多视影像匹配的方法进行研究。多视影像特征点提取和匹配后,使用SFM算法得到对象的稀疏点云数据,因稀疏点云的对象可视化程度低,不能反映对象的特征,因此,在稀疏点云的基础上使用基于片面的多视图立体视觉(patch-based MVS algorithm,PMVS)算法进行密集匹配,获得对象的密集点云数据,能够较完整的反映对象的特征。以密集点云为数据基础,进行下面的点云分类和测图应用研究。(2)多视影像密集匹配获得的密集点云数据为离散点,借鉴LIDAR点云离散点的分类处理算法,重点研究格网化数学形态学法和迭代三角网内插法对倾斜影像密集匹配的点云数据进行处理,通过实验对比两种算法处理的结果,并使用定性和定量评价的方法,分析各算法处理的结果,得出迭代三角网内插法较适用于多视影像密集匹配的点云分类处理,获得地面点和非地面点数据。(3)最后,利用地面点数据和获得的正射图像,进行数字线划图(Digital Line Graphic,DLG)的制作。密集点云数据的密度和数据量较大,处理时比较困难,需要对密集点云数据进行预处理。首先,需要对点云数据进行抽稀处理,然后对抽稀后的简化点云数据进行数字高程模型(Digital Elevation Model,DEM)制作,生成等高线,以正射影像图为底图进行数字线划图的平面图制作,最后,把等高线加入,形成完整的数字线划图。
[Abstract]:With the development of computer vision and photogrammetry, With the use of computer vision motion recovery target structure algorithm (structure form) and multi-view stereotype reconstruction algorithm (multi-view stereotype MVS), the tilt photogrammetry technology for multi-view image matching is developed rapidly. Multi-view tilt photogrammetry can be used to shoot ground objects from many angles, which can obtain the side texture information except vertical direction, and can reflect the texture properties of objects comprehensively. SFM and MVS algorithms are used to match the multi-view images to obtain the sparse point cloud and dense point cloud of the object, and the vertical image is used to make the orthophoto image. The dense point cloud data are classified and processed by grid mathematical morphology method and iterative triangular interpolation method respectively. After classification, the ground point cloud is generated by DEM, and contour lines are generated, and DLG is made by combining with the digitized map of the object of orthophoto image. The methods and results of this paper are as follows: (1) the method of multi-view image matching is studied. After feature points extraction and matching in multi-view images, the sparse point cloud data of objects are obtained by SFM algorithm. Because of the low visualization degree of sparse point clouds, the features of objects can not be reflected. On the basis of sparse point cloud, the one-sided multi-view stereo vision (patch-based MVS algorithm) algorithm is used for dense matching, and the dense point cloud data of the object can be obtained, which can reflect the characteristics of the object completely. Based on the data of dense point cloud, the following point cloud classification and mapping applications are studied. (2) the dense point cloud data obtained by dense matching of multi-view images are discrete points, and the classification and processing algorithm of LIDAR point cloud discrete points is used for reference. This paper focuses on the mesh mathematical morphology method and iterative triangular interpolation method to deal with the point cloud data with dense matching of inclined images. The experimental results of the two algorithms are compared, and the qualitative and quantitative evaluation methods are used. After analyzing the results of each algorithm, it is concluded that iterative triangulation interpolation method is more suitable for point cloud classification with dense matching of multi-view images to obtain ground point and non-ground point data. (3) finally, the ground point data and the obtained forward projection image are used. Make Digital Line drawing (DLG). The density and data amount of the dense point cloud data is large, so it is difficult to process the dense point cloud data, so it is necessary to preprocess the dense point cloud data. First of all, the point cloud data need to be thinned, and then the simplified point cloud data after thinning is made by digital elevation model (Dem), and contour lines are generated, and the plane drawing of the digital line drawing is made with the orthophoto image as the base map. Finally, Add contours to form a complete digital line drawing.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】