多特征结合的倾斜无人机影像匹配方法

UAV Tilted Image Matching Algorithm with Multi-feature Combination

针对无人机倾斜影像存在匹配困难问题,提出融合多种特征优势的无人机影像匹配算法。首先,提取MSER(Maximally Stable Extremal Regions)局部特征稳定区域,并用SIFT(Scale Invariant Feature Transform)描述子对特征进行描述;其次,利用K-D树的搜索策略进行特征点的快速检索,采用NND算法获取初始的粗匹配点对,根据结果计算影像间的仿射变换关系;最后,对SIFT特征点进行约束NCC匹配,利用RANSAC算法剔除外点,完成最终的影像匹配。实验结果表明,该算法对存在较大倾斜角度的无人机影像效果较好,在匹配正确率和仿射不变性两方面都优于SIFT算法。

In view of the difficulty in matching tilted images of UAV,this paper proposes an automatic image matching algorithm for UAVs that combines various feature advantages.Firstly,the local characteristic stable region of MSER is extracted,and the features is described by the SIFT descriptor.Secondly,the K-D tree search strategy is used to quickly retrieve the feature points.The NND algorithm is used to obtain the initial rough matching point pairs.The affine transformation relationship between the images is calculated according to the results.Finally,the SIFT feature points are constrained by NCC matching,the RANSAC algorithm is used.The algorithm excludes the points and completes the final image matching.