摘要
在摄影测量学中应用ORB算法时,影像匹配存在特征分布不均、误匹配率较高及匹配精度为整像素级等不足。针对这些问题进行如下改进,控制特征数目及特征分布,增加核线约束及相关系数条件,结合最小二乘匹配算法。此外,为了增强该算法对大角度旋转影像的适应性,将影像间的仿射变换参数作为最小二乘匹配中几何畸变参数的初始值。实验证明,该算法应用于影像匹配时,不仅可以保持ORB算法的高效性,获得均匀分布、高精度的匹配特征,而且对大角度旋转影像具有一定适应性。
If ORB was applied to feature matching in Digital Photogrammetry, there would be many problems, such as unevenly distributing, high mismatching rate and low matching precision. In this paper, we solved these problems through controlling features＇ number and distribution, appending epipolar constraint and the correlation coefficient, and combining with the least squares matching. In addition, we calculated the affine transformation parameters between images, as initial value of geometric distortion parameters of the least squares matching. Experiments show that this algorithm can not only maintain the efficiency of ORB algorithm, to obtain well-distributed and high accuracy matching features, but also for larger rotation with a certain degree of adaptability.
出处
《测绘地理信息》
2015年第3期31-34,共4页
Journal of Geomatics
基金
国家自然科学基金资助项目(41271454)
关键词
ORB算法
特征匹配
最小二乘匹配
核线约束
相关系数
ORB
feature matching
least squares matching
epipolar constraint
correlation coefficient
