Robust UAV Image Stitching via Multiscale Feature Matching and Local Homography Matrix Estimation

Due to some factors such as optical aberrations and large parallax, a few stitching methods lead to artifacts and distortions in the panoramic image. To alleviate this issue, this paper proposes a robust unmanned aerial vehicle (UAV) image stitching algorithm based on multiscale feature matching and local homography matrix estimation. We firstly employ a multiscale feature matching method to establish a reliable correspondence between two sets of points, so as to suppress artifacts in overlapping areas better. Then, we estimate the local homography matrix to alleviate the distortion of target image by combining similarity transformation of matched feature points with projection transformation. Finally, we use a sigmoid metric to quantify the color differences in the overlapping region and achieve a smooth transition between the images by introducing a seam-cutting method. Extensive experiments demonstrate the effectiveness and robustness of our method for UAV image stitching.