Abstract: This paper presented an optimized algorithm of frame selection on the stitching of videos with overlapped visions. Firstly，it compared the Scale Invariant Feature Transform (SIFT) value differences of all the match points of standard frame and candidacy frames, and then selected the point pairs with minimal SIFT value difference, and calculated the mean values of these differences. Finally, the frame with the minimum mean value was selected as the best frame, and stitched with the standard frame. This algorithm assures the best matching frame and can be used to choose the best stitching frame automatically in video stitching. Simulation shows the effectiveness of this algorithm.

Key words: video stitching

摘要： 提出了一种具有公共视野的视频流拼接中视频帧的选择优化算法，算法计算候选帧与基准帧特征匹配对的特征向量差值的算术均值，选择均值最小的候选帧作为最佳拼接帧，与基准帧进行拼接。实验结果表明，基于该算法拼接的视频流全景视图效果较好，算法很好地解决了视频流拼接中最佳拼接帧的自动选择问题。

关键词: 基于图像的绘制技术, 尺度不变特征变换, 视频流拼接, 帧选择, Image Based Rendering (IBR), Scale Invariant Feature Transform (SIFT), frame selection