POS辅助和核线约束的倾斜影像匹配方法

POS-assisted and nuclear line-constrained oblique image matching method

针对无人机倾斜影像匹配时,由于冗余数据量大、影像几何变形大和重复纹理导致基于SIFT特征点的无人机倾斜影像匹配效率和可靠性低的问题,本文提出一种基于POS辅助和核线约束的倾斜影像匹配方法。在该方法中,首先利用机载GNSS/IMU设备获取的影像POS数据计算影像间在物方的重叠区域,接着将物方重叠区投影至像方,根据两幅影像的像方重叠率筛选高可靠像对;其次采用SIFT-GPU算法对影像提取特征点,并根据POS数据估计像对间的核线关系;然后在核线约束下,以描述子间的欧氏距离为相似性测度,实现特征点的高效稳健匹配;最后采用RANSAC算法剔除误匹配。通过对两组倾斜影像做匹配试验验证了本文方法的可行性。

In order to solve the problem of low efficiency and reliability of UAV oblique image matching based on SIFT feature points due to a large amount of redundant data, large image geometric deformation and repeated texture, a novel image matching method based on POS information and epipolar constraint is proposed. In this method, firstly, the image POS data obtained by the airborne GNSS/IMU device is used to calculate the object overlap area between the images, and then the object overlap area is projected to the image side, and then a highly reliable image pairs are selected according to the image overlap ratio of the two images. Secondly, after using the SIFT-GPU algorithm to extract feature points from the image, the antipolar relationship between the image pairs is estimated based on the POS data. Thirdly, the Euclidean distance between descriptors is a similarity measure, which can achieve efficient and robust matching of feature points. Finally, RANSAC algorithm is used to eliminate mismatches. The reliability of the proposed method is verified by the matching experiment of oblique image.