提出了一种对视频序列进行全景图拼接的方法。主要讨论了有大面积的非刚性运动物体出现的序列，不过此方法也同样适用于无运动物体的纯背景序列。为计算各帧间的投影关系，用仿射模型来描述摄像机运动，并用特征点匹配的方法计算出模型中各参数的值。由于用相关法计算的匹配结果准确率比较低，所以用RANSAC（Random Sampling Consensus）对匹配结果进行了筛选，可以准确求出摄像机运动参数。利用运动参数进行投影，然后用多帧相减并求交集，估计出每帧图像中运动物体存在的区域，最后计算得到了全景图。该方法的结果与前人得到的结果进行了比较，证明用此方法能获得质量较高的全景图。

In this paper we propose a method for making panoramas from video sequence taken by moving cameras.We mainly describe the case where the camera is moving and there are large-area non-rigid moving objects in the scene.But the method also works well when there are no moving objects in videos.Affine model is used to describe the camera motion.To calculate the parameters in affine model,we use feature points matching method.Correlation method is used to match feature points in frames,but correlation method can only give a coarse result which may include some bad matching.To ensure the precision of the motion parameters,RANSAC is used to pick out bad matching.Project frames with motion parameters,and the intersection of two subtraction results of neighbor frames will give the moving object area.Last we can calculate the panorama image.Results given by this method are provided in this paper,and we compare them with other people's results.It proves that our method can make panoramas with high quality.