大视场视频全景图快速生成方法

Fast Construction Method for Large Field Video View Panorama

针对视频图像拼接中配准误差的积累问题,提出一种适合大面积静态场景观测的视频全景图快速生成方法,包括摄像机在一维单幅扫描方式和二维多幅扫描方式下的视频全景图的直接拼接算法。实验结果表明,无论摄像机工作在单幅方式还是多幅方式下,该方法都可以快速、准确地拼接出较高质量的视频全景图像,能有效避免局部配准误差的积累与传播,并解决一般全局配准算法复杂费时的问题。

Aiming at the problem of registration error accumulation, this paper presents a fast video mosaic method especially suitable for real-time observation of large static scene. The proposed method is composed of direct registration techniques for both 1D scan mode and 2D scan mode of a camera. Experimental results show that the proposed algorithm can construct seamless video mosaic quickly both for 1D scan mode and for 2D scan mode of a camera, and it can avoid the error accumulation and propagation resulted from the concatenation of pair-wise transformations. Moreover, the proposed method can greatly reduce the complexity and time cost of the general global alignment algorithms.