【摘要】 航拍图主要被用来作为一些图像处理的基础材料，但是由于无人机飞行高度的自限性和一些干扰因素导致图像的视野和对齐效果受限，致力于研究一种改良的对齐评估方法和运动目标去除以提升全景拼接的效果，以减少局部扭曲，使得拼接结果更加接近自然和平滑.采用网格优化引导拼接来提升全局拼接效果，对于网格的规划采用三个联合的能量函数进行求优，对齐效果的直观感受来源于人类视觉对场景目标中的边缘和线条对齐的观察.因此，加入直线对齐能量函数作为联合的优化方法也能在一定程度上提高对齐的效果，拼接角度问题通过估计一个合适的三维旋转来缓解.最后，对于场景中的动态目标的影响，使用深度学习中的实例分割网络进行潜在的运动目标去除，比如行人、车辆等，并在分割网络输出的目标区域进行图像修复.更多还原

【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.更多还原