基于几何约束与局部描述的航空影像直线匹配

Geometric constraints and local description for aerial images line matching algorithm

针对航空影像场景复杂,干扰直线匹配的不确定因素较多的问题,本文提出了基于几何编组与局部均值标准差直线描述符(MSLD)的直线匹配算法。首先,利用直线检测器(LSD)获得影像的直线特征信息,将直线与直线间的几何特征作为约束条件进行编组,获得两幅航空影像的直线对;其次,运用核线约束确定候选直线对,并依次构建参考与候选直线对的支持区域,利用仿射不变性统一支持区域的大小,借助MSLD进行直线对外观的局部描述;然后,结合欧氏距离与最邻近比值计算航空影像中直线描述符间的相似性,进而确定匹配结果;最后,结合共线约束完成对匹配结果的检核,获取精确的同名直线。试验选择不同场景类型变换下的航空影像数据,结果证明了本文算法可以较好地应对航空影像直线匹配效果不佳的现象。

Due to the complexity of aerial image scene and many uncertain factors that interfere with line matching,we propose a line matching algorithm based on geometric constraint and MSLD in this paper.Firstly,the linear detector LSD is employed to obtain the line feature information of the image.Secondly,according to the geometric features between the different straight lines as the constraint condition,the straight line pair of the two aerial images is obtained by grouping.Then,the candidate line pairs are determined by using the kernel constraints,and the support regions of the reference and candidate line pairs are constructed in turn.The affine transformation is used to unify the size of the support regions MSLD describing the local appearance of lines.By calculating the Euclidean distance between line descriptors in aerial images,candidate lines satisfying the nearest neighbor distance ratio criterion are determined as matching results.Finally,the matching results are checked to remove redundant matching lines.The experiment uses aerial image data of different scene types.The experimental results show that the proposed algorithm can deal with the problem of the poor line matching effect of aerial images.