光束法平差是遥感图像实时拼接系统的关键步骤，它需要消耗大量的运算时间，同时也决定了系统的拼接精度。针对画幅扫描红外成像实时拼接系统中的光束法平差算法，提出了一种实用的光束法平差策略，利用相邻扫描行图像的配准平差，将最新扫描行图像依次配准到已拼接图像中，为拼接图像的实时更新奠定了基础。同时利用GPGPU对光束法平差算法进行了并行加速，在保证精度的情况下，提高了拼接算法速度。通过画幅扫描成像的实时平差实验，验证了算法的精度和实时性。

Bundle adjustment is the key step in real-time mosaicking of remote-sensing imaging. It takes too much calculation time and decides the accuracy of the mosaicking system. Aiming at bundle adjustment algorithm in real-time mosaicking system of frame-sweep infrared imaging, a practical policy of bundle adjustment was proposed. The bundle adjustment between images of two adjacent sweep lines was used to register the lastest sweep line to the mosaicked image, which laid the foundation of real-time updating of mosaic image. The GPGPU platform was used to implement and accelerate, which improved the real-time performance of mosaicking system and also guaranteed the accuracy. By implementation of real-time bundle adjustment of frame-sweep images, the precision and real-time performance of the bundle adjustment algorithm were validated.