像方与物方相互结合的无人机航空影像匹配研究

Research on multi-view image matching of UAV aerial image

提出了将像方与物方相结合的两阶段的无人机航空影像匹配方法,该方法可兼顾多视影像匹配可靠性和高精度空三解算的要求,同时也顾及了匹配效率。该方法利用无人机影像航带内重叠度大,影像间的变形对匹配影响小的优势,借助循序渐进的思路,在'像方'使用缩小影像快速开展多视影像匹配,同时采用光束法平差解算影像较为精确的外方位元素和计算地形数据(DSM)并对影像正射纠正。接下来,在'物方'利用正射影像开展二次匹配,这样不仅可以减少影像旋转、尺度缩放、放射变换、投影变形等对匹配的影响(特别是减小投影变形对航带间影像匹配的影响),提高了多视影像间匹配同名点的数量及分布均匀度,增强匹配的可靠性;在'物方',还保证了SIFT算法在同一影像尺度下提取特征点,提高了空三解算精度。通过选取不同地形类别的无人机航空影像试验验证,表明在'物方'采用正射影像开展多视影像匹配,航带间影像匹配点的数量明显提高(丘陵等平坦区域,点的数量增加约6倍;高山区域,点的数量增加约4~6倍,其中重叠度较小的影像间增加约4倍,重叠度较大的影像间增加约6倍),较传统的SIFT算法,空三解算精度提高约30%~40%。

In this paper,the matching of the image space and the object space is combined to improve the reliability of multi-view image matching.First,in the image spaceusing the advantages of the large forward overlap,we use the abbreviated image to matchthe corresponding points,and adjust the exterior orientation element of image,and make sparse DSM(Digital Surface Model),and rectificate orthophoto for each image.This not only removes the deformation of the image rotation,affine transformation,etc,but also reduces the influence of the height displacement caused by the terrain.Next,use the orthophoto in the object space to match corresponding points.This method not only improves the number of matching correspondingpoints,but also enhances the uniformity of the distribution of the matching corresponding points.