Research on low-altitude aerial panoramic image stitching

Dai Jiayin,Wang Yuxin,Yuan Jie

DOI: https://doi.org/10.13232/j.cnki.jnju.2023.02.006

2023-01-01

Abstract:Aerial photos are used as the basic material for some image processing. However,due to the self-limitation of the drone’s flying height and some interference factors,the field of view and alignment of the image aren’t good. This paper is devoted to studying an improved alignment evaluation method and moving target removal to improve the effect of panoramic stitching,reduce distortion,and make the stitching more natural and smooth. Here,grid optimization is used to guide stitching to improve the global stitching,for this method,three joint energy functions are used for optimization. The intuitive feeling of the alignment effect comes from the human vision’s observation of the edge and line alignment in the scene,therefore,adding the linear alignment energy function as a joint optimization method can improve the alignment effect to a certain extent. And the angle problem is alleviated by estimating a suitable 3D rotation. Finally,for the impact of dynamic objects in the scene,this paper uses the instance segmentation network in deep learning to remove potential moving objects,such as pedestrians,vehicles,etc.,and performs image inpainting in the target area output by the segmentation network.