中文摘要：

在传统的视频拼接算法中,配准的误差和运动的物体都会使拼接结果产生鬼影,而复杂的融合算法又难以满足实时性要求。针对上述问题,以平行光轴且光心位于同一水平基线位置的双目相机采集的视频图像为研究对象,提出一种基于限定区域和SURF算法的视频拼接方法。针对采集到的具有20%~30%重叠感兴趣区域（ROI）的视频图像,每24帧设置一个关键帧,仅对关键帧的ROI区域采用SURF算法寻找特征匹配点对。然后采用改进的RANSAC算法筛选单应性变换矩阵H,对非关键帧直接采用此单应性变换矩阵H进行图像融合。实验结果表明,采用关键帧计算特征点的方式得到的视频拼接效果能够很好地消除鬼影,同时也能够保证视频融合的实时性。

英文摘要：

In the traditional video stitching,the calibration error and the motion of an object may cause the ghost shadows in stitching results. In addition,the complex fusion algorithm can＇t meet the real-time requirement. In order to solve the above problems,the video images acquired by binocular camera that has a parallel optical axis and its optic center is at the same horizontal baseline are studied,and a new method of real-time video stitching based on limited region and SURF algorithm is put forward in this paper. A key frame is set per 24 frames for the video images with 20%~30% overlap area in ROI area. the SURF algorithm is used to look for the feature matching point pairs in the ROI overlap area of key frame only. The homography transformation matrix H is selected with the improved RANSAC algorithm,and then the image fusion is executed for non-key frames by using the transformation matrix H directly. The experiment results show that the video stitching algorithm proposed in this paper can eliminate the ghost shadows well,also can ensure the real-time performance of video.