与普通场景图像相比,无人机影像中纹理信息较丰富,局部特征与目标对象"一对多"的对应问题更加严重,经典SURF算法不适用于无人机影像的特征点匹配。为此,提出一种辅以空间约束的SURF特征点匹配方法,并应用于无人机影像拼接。该方法对基准影像整体提取SURF特征点,对目标影像分块提取SURF特征点,在特征点双向匹配过程中使用两特征点对进行空间约束,实现目标影像子图像与基准影像的特征点匹配;根据特征点对计算目标影像初始变换参数,估计目标影像特征点的匹配点在基准影像上的点位,对匹配点搜索空间进行约束,提高匹配速度与精度;利用点疏密度空间约束,得到均匀分布的特征点对。最后,利用所获取的特征点对实现无人机影像的配准与拼接,通过人工选取均匀分布的特征点对验证拼接精度。实验结果表明,采用本文方法提取的特征点能够得到较好的无人机影像拼接效果。

Compared with general scene images, unmanned aerial vehicle(UAV)images provide richer texture information, and often there are more serious problems for the one-to-many correspondence between local features and target objects. The traditional speeded-up robust features (SURF) algorithm would be inapplicable to UAV images. Therefore, an improved spatially constrained SURF method is proposed for UAV image matching and mosaicking. In the first feature point matching phase, SURF feature points are extracted from the whole base image and the blocks of the target image, respectively, a cosine-based spatial constraint relationship is built using the selected two pairs of points and imposed on the feature point dual matching process between the central block in the target image and the base image. In the second phase, the initial parameters of geometric transformation are calculated using the feature point points obtained in the first phase and used to estimate locations in the base image corresponding to points in the target image. For each feature point in the target image, point matching just need to be done within the neighborhood of the estimated locations so as to ensure matching efficiency and reliability. Meanwhile, uniformly distributed feature points are achieved with the constraint of point intensity. Finally, the obtained feature points are used for UAV image registration and mosaicking. The performance is compared with the manually selected points that are uniformly distributed. Experimental results illustrate the validity of the presented method.