结合Harris-Laplace与SIFT特征的无人机倾斜影像拼接

Tilting Image Mosaic of UAV Combined with Harris-Laplace and SIFT Features

针对无人机影像存在仿射变形与阴影问题,本文提出应用Harris-Laplace与SIFT特征的倾斜无人机影像匹配方法。首先,提取具有光照、影像噪声、尺度不变性的Harris-Laplace关键点,并计算关键点的主方向,生成特征点;然后采用SIFT特征描述子对第一步提取的特征点进行表达;最后,采用BBF方法提取初始匹配点对和最小二乘法约束的均方根误差(RMSE)剔除。实验结果表明,该算法在存在仿射变形、高大建筑物阴影的影像匹配表现较好的结果。

Aiming at the problem of affine deformation and shadow in Unmanned Aerial Vehicle（UAV）image,this paper proposes a matching method of tilting UAV image based on Harris-Laplace and SIFT features.First,the key points of Harris-Laplace with light,image noise and scale invariance are extracted,and the main direction of the key points is calculated and the feature points are generated.Then the feature points extracted from the first step are expressed with the SIFT feature descriptor.Finally,the BBF method is used to extract the mean square of the initial matching point pair and the least square method.Root error（RMSE）is eliminated.Experimental results show that the algorithm performs well in image matching with affine deformation and tall building shadow.