一种内容完整的视频稳定算法  

Video stabilization algorithm content complete

作　　者：彭艺[1] 叶齐祥[1] 黄钧[1] 焦建彬[1] 

机构地区：[1]中国科学院研究生院,北京100049

出　　处：《中国图象图形学报》2010年第9期1384-1390,共7页Journal of Image and Graphics

基　　金：国家自然科学基金项目(60672147;60872143)

摘　　要：设计了一种基于可靠特征集合匹配的内容完整的视频稳定算法。为了避免运动前景上的特征点参与运动估计,由经典的KLT(Kanade-Lucas-Tomasi)算法提取特征点,而后基于特征有效性判定规则对特征点集合进行有效性验证以提高特征点的可靠性。利用通过验证的特征点对全局运动进行估计,得到精确的运动参数并据此对视频图像进行运动补偿。对于运动补偿造成的无定义区,首先计算当前帧的定义区与相邻帧的光流,以此为向导腐蚀无定义区;利用拼接的方法,填充仍为无定义区的像素。实验结果表明该算法对于前景物体运动具有较好的鲁棒性并能够生成内容完整的稳定视频序列。Although there are many feature matching and tracking methods so far, the side effect of moving foreground object, which will cause global motion estimation error, is still an open problem. In order to avoid features, located on the foreground objects, participating in motion estimation, feature effectiveness evaluation is employed to improve feature reliability for the features extracted by the traditional KLT method. Effective features are utilized to estimate global motion and obtain accurate motion parameter, based on which video frames are compensated. However, motion compensation will cause undefined area. There are some approaches to reconstruct the undefined area; nevertheless, they have not considered the effect of fast moving objects in the foreground of the video, which will decrease the video quality after stabilization and content completion. In our proposed algorithm, optical flow between defined areas of current frame and neighbor frame is first calculated, and then it is used as a guide to erode unknown areas. Finally, mosaicking on the base of reference frame is used to obtain a complete video stabilization sequence. Experiment results show that the proposed method is robust to moving foreground objects and is able to realize video frames stabilization with complete content.

关 键 词：视频稳定 KLT算法 光流 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]