Constructing high-quality panorama is a fundamental task in both computer vision and computer graphics communities, thus, leading to many image stitching approaches for panorama construction. However, panorama constructed by traditional image stitching has a limited angle of view and has an expensively computational cost. To defend the issues, in this paper, we proposed to use video sequence as input for constructing big panorama, resulting in a high-quality panoramic image, the presented method is built on the stabilized video and robust feature tracking method. Specifically, the input video sequences are captured by moving hand cameras which can be any type of consumer-level camera. To mitigate the affections from rolling shutters in videos, a novel video stabilization method is introduced to filter the unstable camera's path, then resulting in a stabilized video for panorama construction. Additionally, a deep local feature-based feature tracking method is proposed to produce feature correspondences between consecutive video frames for camera motion estimation used in both video stabilization and image stitching. Finally, a comprehensive experiment conducted on the benchmarking datasets is presented to demonstrate the effectiveness of the proposed method.

在计算机视觉和计算机图形社区中，构建高质量全景图都是一项基本任务，因此，导致了许多用于全景图构建的图像拼接方法。然而，通过传统的图像拼接构造的全景图具有有限的视角并且具有昂贵的计算成本。为了解决这些问题，本文提出以视频序列作为输入来构建大全景图，从而得到高质量的全景图像，该方法建立在稳定的视频和鲁棒的特征跟踪方法的基础上。具体而言，输入视频序列是通过移动手持摄像机捕获的，移动手持摄像机可以是任何类型的消费者级别的摄像机。为了减轻视频中滚动快门带来的影响，引入了一种新颖的视频稳定方法来过滤不稳定的摄像机路径，然后生成稳定的视频以进行全景制作。此外，提出了一种基于深度局部特征的特征跟踪方法，以在视频稳定和图像拼接中使用的摄像机运动估计的连续视频帧之间生成特征对应关系。最后，提出了对基准数据集进行的综合实验，以证明该方法的有效性。