相机阵列中无视场重合图像时间配准方法研究

多路商业相机组阵在进行信息融合与视场拼接之前需要对采集的多路图像进行时间配准。为保证通用性，需要研究一种与系统相对独立的多路图像时间配准方法。提出一种利用多路图像序列内变换关系的算法，首先提取多路图像中与时间相关的共同特征，然后通过改进的奇异值分解方法判定共同特征的相似度完成时间配准。本文算法不需要提前对相机进行标定，对序列内图像配准精度要求较低，且能够在多路图像之间没有视场重合的情况下完成时间配准。经实验验证，配准精度达到了亚帧级。

Method suitable for image registration of camera array used in space target observation

Increasing commercial camera resolution and detection sensitivity make it possible to observe the target space, And adopt the method of multiple unit array can further expand its field of view. Before the information fusion there must be a time registration to the acquisition of multi channel image; because the storage medium, such as CF card, the time registration can only use the relevant information in the image by post processing. This paper proposes a method by using multiple image sequence in transformation relationship, which can finish the time registration without the criteria that each image must of the same view. And it can effectively overcome the effects of noise and error matching, verified by the experimental registration accuracy can reach millisecond level.