【作者】 许彪；

【导师】 张剑清； 张勇；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2012， 博士

【摘要】 数字正射影像(DOM)作为一种重要的基础地理信息产品，兼备丰富的纹理信息和地图的几何属性，其制作技术在过去几十年中得到了极大的发展。一方面、DOM制作的主要目的在于消除地形起伏和相机倾斜等引起的投影变形，正射纠正过程中利用的是定向后的影像和数字高程模型(DEM)，并未考虑如人工建筑物等空间目标，从而导致这些目标的中心投影变形依然存在。另一方面、随着数码相机成像技术的不断提高，目前已经能够获取小于0.1米分辨率的航空影像，例如建筑物、桥梁等人工地物目标在影像上清晰可见。正射纠正时如果不对这些目标加以考虑，势必会导致DOM中出现建筑物倾斜、遮挡等问题，如果将GIS矢量地图叠加到DOM上，会表现出道路矢量线横穿建筑物以及建筑物矢量轮廓无法与其自身套合等问题，严重时将会导致DOM失去地理参考价值，不足以作为基础底图使用。真正射影像(True Orthophoto)即是在上述背景下提出来的一种更高级的正射产品。真正射影像中如建筑物、桥梁等人工目标能够纠正到其正确的位置上，不会对其他地物目标(如道路)造成遮挡，真正意义上兼备了丰富的纹理信息和地形图的几何属性，受到了国内外摄影测量学者的广泛关注。真正射影像与DOM最显著的差异在于正射纠正的同时分析地物的可见性。可见区域的纹理获取与制作DOM相同，即灰度内插和赋值；遮蔽区域的纹理利用相邻影像中的可见部分予以修补。因此，制作真正射影像中最为关键的技术环节为可见性分析，也可以称为遮蔽区域检测，各种真正射影像制作方法的本质差异就在于遮蔽检测方法的不同。从国内外研究成果上看，遮蔽检测主要利用数字表面模型(DSM)和数字建筑物模型(DBM)。鉴于上述分析，本文的研究目的即为如何通过影像匹配的方式自动获取DSM，并利用DSM或DBM检测遮蔽区域，结合相邻影像的可见部分纹理修补遮蔽区域用以制作真正射影像，主要研究内容包括以下方面：顾及面特征匹配的数字表面模型获取算法。面特征相比较点、边缘等特征，具有区域闭合的特点，将面特征匹配引入到多影像多基元的匹配算法中，一方面、面特征的匹配可以为面特征边界点及其他点(特征点、格网点、边缘关键点等)的匹配提供初值，面特征边界点自身的匹配结果具有很高的可靠性，可作为点、线混合概率松弛法全局一致性匹配的“锚点”使用；另一方面、多种匹配基元的相互融合，可以获取数目巨大的匹配特征，有助于获取密集、精确及可靠的数字表面模型；基于DSM的遮蔽区域检测。改进一种现有的基于DSM的遮蔽区域检测算法，本文称为基于高程约束的遮蔽检测算法，该方法以摄影中心对应的地底点为起始点同时假定地底点附近4个DSM格网点可见，采用螺旋扫描的方式自内向外依次检测DSM格网点的可见性。基于DBM的遮蔽区域检测。提出一种基于三角网的遮蔽检测算法，该方法以单个三角形为检测单元，通过反演中心投影成像时的视觉状态，对三角形进行视觉排序，依次将三角形反投影到像方，同时合并所有像方投影区域构建投影多边形，根据三角形像方投影区域与投影多边形关联情况判断视觉状态。真正射影像制作。采用“权值影像“的方式结合相邻影像的可见部分纹理修补遮蔽区域。“权值影像”是指对可用于修补主影像遮蔽区域的所有副影像计算一个质量评价标准，即权值，选择最大权值对应的副影像作为修补源影像。更多还原

【Abstract】 Digital orthophoto(DOM) as an important basic geographic information products thathas both richly textures and geometric properties of the map, and its technology of productivehas been greatly evolved over the past few decades. The main purpose of DOM generationaims at eliminating projection distortion of terrain and camera tilt. We know thatorthorectified using for both orientation images and Digital Elevation Model(DEM), notinclude spatial object such as buildings that still exist the problem of projection distortion. Onthe other hand, with the development of digital camera technology, it has been able toacquired aerial images that resolution is less than0.1m, it means that the spatial object such asbuildings and bridges is very clearly in the images. The problem of both buildings tilt andocclusion will be yet existed if we are not considering these objects when orthorectifiedprocessing. As a result of there will be shown two mainly contradictions if the vector maps ofGIS are superimposed on to the DOM, firstly the buildings are acrossed by vector lines ofroads, and secondly vector outlines of the buildings cannot be fitted to itself. It will be loss thevalue of geo-referenced and cannot be used as a basic map if the problems of abovementioned are more seriously.True Orthophoto belongs to more advanced ortho products, and it is proposed in theabove context. The buildings and bridges are rectified to correct position in true orthophoto,and it will not be occluded other objects(such as roads). Both richly textures and geometricproperties of the map are really possessed by true orthophoto, so it has been widely concernedby lots of scholars of photogrammetry.The most significant difference between true orthophoto and DOM is visibility analysiswhen orthorectified processing. The textures of visible region are provided by grayinterpolation and assignment that is equivalent to tradition method of rectification. Thetextures of occluded region are repaired by visible parts of neighbour images. The visibilityanalysis is the most significant technology in the true orthophoto generating, it is also calledocclusion area detection. Now, the methods of visibility analysis that mainly making use ofDigital Surface Model and Digital Building Model are different according to various methodsof true orthophoto generation. As a result of this paper mainly aims at automatically obtainingDSM through image matching, and detecting occlusion areas through DSM or DBM, andgenerating true orthophoto by means of textures repairing.The main research contents areintroduced as follows:The matching algorithm of DSM is taken into account feature planars that have regional closure compared to feature points and edges, which is called Multiple Primitive Multi-ImageMatching. The matching of the feature planars can provide approximate value using formatching others points include feature points, grid points, key points of edges and boundarypoints of feature planars. The matching results of the boundary points of feature planars havevery high reliability, there can be served as ‘anchor’ using for point-line hybird globalconsistency matching with probability relaxation technique. On the other hand, a hugenumber of matching features through merging multiple matching primitives have ability toprovide dense, precise, and reliable results.An existence method of occlusion detection based on DSM is improved, and it has beenknown as Occlusion Detection based on Elevation Constraint. This method assumes fourDSM cells that are surrounding the nadir points are visible, and it detects the visibility of theDSM cells starting from the nadir point in a spiral mode from inside to outside.The occluded regions can be detected with the DBM, so-called Occlusion Detectionbased on TIN. The triangle is the detection unit in this method. Firstly, visual sorting oftriangles are primarily processed through inversing visual status of central projection.Secondly, it successively projects the triangles of sorted to image space, and projectionpolygons of image space can be constructed through merging all projection areas of triangles.The relationship between projection areas of triangles and projection polygons can be used fordetecting visual status.True orthophoto generation combines master image and slave images, and occludedregions are repaired through ‘Weight Image’, it means a quality measures in a so-calledweight, which is computed for each individual available slave images. Texture repair issimply picking the slave image with the highest weight.更多还原