【作者】 王宁；

【导师】 张卡；

【作者基本信息】 南京师范大学 ， 地图学与地理信息系统， 2017， 硕士

【摘要】 通过高分辨率遥感影像立体像对生成三维地形是摄影测量领域一个重要方向,其中影像匹配是其核心环节。我国已有多颗在轨运行的高分辨率遥感卫星,但有时由于技术条件或者外界条件限制,可能无法获得同一地区的同源立体像对。因此异源影像匹配方法已然成为影像匹配领域的研究热点,本文的异源影像是指资源三号不同时相或者不同视相的影像。由于异源影像具有几何变形不一致,辐射差异大,空间分辨率不同等特点使得异源影像匹配较困难。本文详细研究了异源影像RFM模型的核线偏差,并在此基础上选用基于ZNCC相似性测度以及核线约束的匹配方法和基于种子点的匹配方法进行密集匹配并比较两者结果,最后对密集匹配点使用RFM前方交会生成南京某地区的地面密集点云。本文的主要结论有:(1)在分析异源影像特点的情况下,为了使两幅待匹配的影像质量较为一致,本文对试验区影像进行直方图匹配后再进行同名点提取,大大提高同名点提取的质量与数量。(2)本文研究了多种稳定同名点提取的方法,最后选用基于SIFT的特征匹配算法提取异源影像稳定同名特征点并进行匹配,之后对匹配后的点进行比率剔除和RANSAC剔除,确保了同名点的准确性。(3)研究了 RFM反解模型的计算方法,使用投影轨迹法核线模型和稳定同名点定量化研究了线阵影像同名点与同名核线的偏差。并指出使用同一视角异源影像计算的核线偏差小于使用不同视角的异源影像计算的核线偏差。原因可能是同一视角传感器几何畸变较为一致,RPC系数解算时的精度也较为一致。(4)本文使用资源三号2013年11月4号庐山地区后视和下视以及南京地区2014年3月17号和2014年3月22号异源影像进行稳定同名点提取实验,并在此基础上计算了核线偏差,统计核线偏差在行列方向的值。此外本文使用南京地区3月17和22号后视一块影像的同名点匹配结果生成该区域的密集匹配点云。最后与基于种子点搜索法生成的密集点云做对比分析。更多还原

【Abstract】 The construction of 3-D terrain model using high resolution remote sensing images is an important direction in the field of photogrammetry. Image matching is the core of this process. There are many high-resolution remote sensing satellites orbiting in China, however, sometimes due to technical or environmental restrictions,it is difficult to get the same-source images in the same area. Therefore, multi-source image matching has become a research trend in the field of image matching recently.As the geometric and radiometric deformation of multi-source stereoscopic Images,which makes the multi-source image matching extremely difficult. Different views of ZY-3 images was used to illustrate methods proposed here. In this paper, the deviation of the epipolar error of the ZY-3 multi-source images was carefully studied. On the basis of this, The searching area of matching can be limited to epipolar nearby. Finally a dense matching method based on ZNCC combined with epipolar constraint was tested, whose result was compared with the matching method based on seed pixel,then RFM forward intersection was used to generate ground dense points in a region of Nanjing.The main conclusions of this paper are:(1) In order to make the quality of multi-source images consistent, histogram matching method was used to improve the quality of the bad images, this method can greatly improve the quality and quantity of the extracted tie-points.(2) The algorithm of many points-extract method was tested. SIFT was choosed to extract feature points, then RANSAC was used to eliminate the wrong tie-points,guaranteeing the points are right.(3) In order to improve the efficiency of dense matching method, the calculation of inverse RFM model was studied. The deviation between tie-points and epipolar was also calculated. In the process of our experiment, finding the deviation is correlated with the image type and imaging angle of the research area.(4)The back-view and nadir-view image in Lushan on 4 Nov, 2013 and the multi-source image in Nanjing on 17 Mar and 22 Mar, 2014 was used to test the SIFT based matching method. These extracted tie-points was used to calculate the epipolar error.Finally, the back-view image on 17 Mar, 2014 and 22 Mar, 2014 in Nanjing was used to generate dense ground points based on ZNCC similarity measure更多还原