A GPU Parallel Stitching Algorithm for Low-Altitude Parallax Scenes

Panorama stitching of low-altitude captured images is a very interesting and challenging task. Since 3D scenes are not planar, there may be significant changes in the relative positions of scene structures in each perspective in the images, thus causing parallax. Existing image stitching algorithms often fail in parallax scenes and are not efficient enough to meet the real-time processing requirements of drones. To address this problem, we introduce gridded dense projection estimation and design a parallel computing unit suitable for GPUs for rapid modeling of projection fields. Experimental results show that our algorithm significantly outperforms the comparison algorithms in terms of computational performance, has high real-time performance, and achieves nearly the same level of stitching quality and naturalness.