Abstract: A novel algorithm for video mosaicking with high accuracy is proposed to achieve the requirement of complex dynamic scenes based on contours feature of main regions.This algorithm overcomes the drawback of low accuracy of traditional methods caused by complicated scene transformation or scene of low overlap.Experiments on real video sequences show that this algorithm can get high quality result in accuracy and is not only more suitable for affine transformation，but also robust in processing the video with low overlapping area and occlusion compared with mosaicking algorithm based on SIFT.

Key words: video mosaic, matching, contours

摘要： 为了解决复杂场景变换时视频拼接的精度问题，提出了一种仿射不变轮廓匹配的拼接算法——利用场景中主要区域的轮廓信息对视频图像进行匹配，然后利用匹配点求解运动参数进行拼接。该方法克服了传统拼接算法在复杂的场景变化以及重叠区小的情况下拼接精度低的问题。与基于SIFT的拼接算法对比表明，该算法实现了图像序列的高精度拼接，不仅能适应仿射变换，并且对于重叠区域小，存在运动遮挡的视频具有一定的鲁棒性。

关键词: 视频拼接, 匹配, 轮廓