基于欧氏距离变换的无人机图像拼接

Drone Image Stitching based on Euclidean Distance Transform

针对无人机航拍图像线性拼接算法不适用于不规则融合区域的问题,提出了一种基于欧氏距离变换的无人机航拍图像融合算法。先通过加速鲁棒特性(Speeded-Up Robust Features,SURF)算法进行图像匹配,再采用欧式距离变换法来实现在重叠区域的加权融合。实验结果表明,相对于传统的线性加权融合算法,该方法显著消除了融合过程中出现的拼接痕迹和鬼影问题。对于航拍图像拼接等重叠区域不规则、重叠位置多变的多图像拼接问题,该算法拥有良好的拼接效果。

In drone aerial image stitching problems, stitching boundaries and ghosting phenomena often appear in the traditional linearly weighted image fusion method. Therefore, an image stitching algorithm based on Euclidean distance transform is proposed in this paper. Firstly, the image features are extracted by SURF(Speeded-Up Robust Features) algorithm. Then, a rough matching of feature points is produced by KNN algorithm before RANSAC method is used to eliminate matching errors. After image warping, Euclidean distance transform is applied to construct pixel weights of unwrapped and warped images on the overlapping regions, which are then multiplied to corresponding images and added together to achieve image fusion. Experimental results indicate that the proposed method could remarkably diminish seams and ghosting effects in stitching processes, comparing to the traditional linearly weighted stitching algorithm. This method also produced decent outcomes when applied to drone images, which have irregular overlaps at random positions.