一种分层描述符的无人机影像匹配方法

A layered descriptor based unmanned aerial vehicle image matching method

针对无人机航摄由于对地摄影,地面、河流、居民区等地物地貌容易出现许多弱纹理与重复纹理区域,传统影像匹配算法针对此容易出现鲁棒性不足的问题,该文提出一种基于均匀ORB特征与格式塔心理学邻近原则分层二进制描述符的改进ORB算法。该方法先对ORB角点进行均匀化操作,以提升特征分布质量;基于格式塔心理学邻近原则将大特征邻域划分为“重要”与“一般”两层区域,以构建可区分性更高的描述向量参与匹配。实验表明,该方法在含有较多弱纹理或重复纹理区域的影像上正确匹配分布质量与平均匹配正确率上均优于ORB、BRISK与SIFT算法,平均匹配正确率分别高约11.10%、5.68%与11.98%;且能在更多的弱纹理与重复纹理区域获得正确匹配。

Due to the aerial photography of unmanned aerial vehicles,many weak and repetitive texture areas are prone to appear on the ground,rivers,residential areas,and other terrain features.Traditional image matching algorithms are prone to insufficient robustness in this regard.In view of this,this paper proposes an improved ORB algorithm based on uniform ORB features and Gestalt psychology proximity principle hierarchical binary descriptor(GED).This method first performs a uniform operation on ORB corners to improve the quality of feature distribution;Then,based on the proximity principle of Gestalt psychology,the large feature neighborhood is divided into“important”and“general”two-layer regions to build a more distinguishable description vector for matching.Experiments have shown that this method outperforms ORB,BRISK,and SIFT algorithms in accurately matching distribution quality and average correct matching rate on images with many weak or repeated texture regions,with average matching accuracy rates of about 11.10%,5.68%,and 11.98%higher,respectively;And it can achieve more correct matching in weaker texture and repeated texture regions.