倾斜影像与垂直影像的逐层自动密集匹配  

Layer-by-layer Automatic Dense Matching of Oblique Image and Vertical Image

作　　者：张丽[1,2] 龚辉 纪松[3] 张振超 

机构地区：[1]西安测绘研究所,陕西西安710054 [2]地理信息工程国家重点实验室,陕西西安710054 [3]信息工程大学地理空间信息学院,河南郑州450052

出　　处：《测绘科学与工程》2017年第6期41-46,共6页Geomatics Science and Engineering

摘　　要：针对多视角影像中，垂直影像和倾斜影像分辨率差异大、变形严重、尺度不统一造成匹配困难的难题，提出了一种基于三层金字塔的自适应密集匹配方法。首先在金字塔顶层，采用ASIFT算法进行同名特征点的提取；其次利用顶层传递来的特征点对中间层影像进行等效下视纠正，并利用SURF算法匹配同名特征点；最后利用传递来的特征点对原始影像进行基于三角网的局部纠正和基于灰度的密集匹配。实验结果表明，匹配精度优于1个像素的点数超过了75％，可为多视角影像整体点云的拼接与融合提供控制。In order to solve the matching problem of muhi-perspective images caused by the great disparity of image resolu- tion, serious deformation and non-uniform scale between vertical image and oblique image, an adaptive dense matching method based on three layer pyramid is proposed in this paper. Firstly, ASIFT algorithm is used to extract corresponding feature points in pyramid top layer. Secondly, feature points transferred from the pyramid top layer are used to correct the middle layer image as the equivalent vertical image, and SURF algorithm is used to match the homologous feature points. Finally, the transferred feature points are used to do local area rectification based on triangulation and dense matching based on image gray scale. Experiment re- sults show that the matching precision of 75% points are better than 1 pixel,which can help to control the mosaic and integration of the whole point cloud of multi-perspective images.

关 键 词：多视角影像 倾斜影像 垂直影像 密集匹配 ASIFT SURF 

分 类 号：P231[天文地球—摄影测量与遥感]