边缘约束的航摄立体影像密集匹配方法  

Edge-constrained Dense Matching Method for Aerial Stereo Images

作　　者：纪艳华 许殊 陈勃 JI Yanhua;XU Shu;CHEN Bo(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094,China)

机构地区：[1]武汉大学遥感信息工程学院,湖北武汉430079 [2]中国科学院空天信息创新研究院,北京100094

出　　处：《测绘地理信息》2022年第2期87-91,共5页Journal of Geomatics

基　　金：中国科学院战略性先导科技专项(A类)(XDA19000000)。

摘　　要：航摄影像中的房屋边缘等深度不连续给密集匹配带来了极大的挑战。传统半全局密集匹配(semi-global matching,SGM)方法中使用双参数进行视差平滑约束,但对整张图像使用同一参数难以适应航摄影像中复杂的场景;参数太大时易导致深度不连续处过度平滑,丢失边角特征,参数太小时易导致在平坦区域平滑约束不足,造成点云空洞。鉴于以上问题,本文提出了一种边缘约束的航摄立体影像密集匹配方法:首先利用影像的点互信息构建相似性矩阵从而得到每个像素的边缘概率;其次,根据像素的边缘属性为每个像素选择不同的惩罚参数;最后根据上述选择的惩罚参数进行半全局密集匹配。通过在Vaihingen数据集进行试验分析证明,本文方法的点云结果不仅在平坦区域保持完整,同时更好地保留了房屋的边角特征。Depth discontinuities such as the edges of houses in aerial image bring great challenges to dense matching.Two parameters are used in the traditional semi-global dense matching method for disparity smoothing constraints.But it is difficult to adapt the same parameters to the complex scenes in an aerial image.larger parameters will impose excessive constraint on smoothing at the disparity discontinuities and result in losses of sharp features.smaller parameters will impose insufficient smoothing constraints in the flat area and result in point cloud vulnerability.In view of the above problems,this paper proposes edge-constrained dense matching method for aerial stereo images.Firstly,the similarity matrix is constructed based on the point mutual information of the image,and then the edge information of each pixel is obtained.Secondly,the penalty parameters for each pixel are selected according to the edge properties.Finally,semi-global dense matching is performed according to the penalty parameters selected above.Experimental evaluations using the Vaihingen dataset have revealed that results of the proposed method not only remain intact in flat areas,but also retain the sharp features better.

关 键 词：航摄影像 密集匹配 边缘信息 惩罚参数 

分 类 号：P231[天文地球—摄影测量与遥感]