Abstract: This paper provides a method stitching real-time video images captured by several cameras.The algorithm is based on SIFT（Scale Invariant Feature Transform） feature tracking in video frames and reduce the cost of searching in for matches.SIFT is used to extraction feature descriptor of frames and matched in tracking estimating area，then estimating the transformation parameters of the frames stitching.Experimental results show the method is wonderful and stable.It is also robust even in processing low overlapping area，any relative orientation and between the videos frames or moving objects.

Key words: SIFT feature matching, image registration, video stitch, image mosaic

摘要： 针对不同摄像头的监控视频序列，提出了一种基于视频帧SIFT（Scale Invariant Feature Transform，即尺度不变特征变换）特征跟踪的拼接方法。通过SIFT算法提取帧图像的特征，并在跟踪的估计区域搜索匹配特征，从而确定待整合帧之间的变换参数。实验结果表明，该方法较好实现视频快速拼接，且对重叠区域小、形变大、有运动物体遮挡的视频具有较强的鲁棒性。

关键词: SIFT特征匹配, 图像配准, 视频拼接, 图像拼接