A Fast and Effective Panorama Stitching Algorithm on UAV Aerial Images

Abstract

Nowadays, UAV(Unmanned Aerial Vehicle) has been widely used in many areas, such as agriculture, rescue and disaster relief, video shooting. In some applications, the whole environment needs to be perceived, so image stitching comes in handy. It is time-consuming for present methods to stitch aerial images, whose bottleneck is to generate dense point clouds. Besides, traditional image stitching algorithm may encounter accumulative error which makes the result distortion. In this paper, a fast UAV image stitching algorithm is proposed by re-placing generation of dense point clouds with image composition algorithm and taking full advantages of GPS information to initialize cameras' rotation to weaken the impact of accumulative error. The proposed model is much simpler than classical models in UAV image stitching, it can handle large scene very quickly. Furthermore, the effect of stitching is better than the current commercial software(Pix4D). When overlap ratio between two images is low, the proposed algorithm can generate a good stitching result, while commercial software may generate some black areas.