一种抗信息丢失的稳像系统  

A video stabilization system for preventing information loss

作　　者：陈晓露[1,2] 戴明[1] 尹传历[1] 王子辰[1,2] 孙乐[1,2] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所中国科学院航空光学成像与测量重点实验室,吉林长春130033 [2]中国科学院大学,北京100039

出　　处：《光电子．激光》2013年第7期1416-1421,共6页Journal of Optoelectronics·Laser

基　　金：国家“973”重点基础研究发展规划(2009CB72400603B);国家“863”高技术研究发展计划(2008AA121803)资助项目

摘　　要：针对二维电子稳像补偿全局运动矢量后会出现大量的空白区域,提出了一种不需要采用单应性模型明确估计全局运动矢量的快速平滑特征轨迹的稳像算法。首先,采用改进的快速鲁棒特征(SURF)提取图像局部特征点;然后,利用空间运动的一致性连接帧与帧之间匹配的特征点得到特征点轨迹;最后,建立同时考虑特征轨迹的平滑度和视频质量退化程度的目标函数平滑特征点轨迹,得到稳定的视频。实验结果表明,用本文方法稳定的视频比Matsushita方法处理后的视频丢失的区域减小了30%左右,更满足人眼感官需求,减轻了费时的运动修复任务;同时消除了运动估计中帧间匹配的累积误差,对前景存在较大局部运动的视频仍能表现较好的稳像效果。To solve the problem that there will be massive blank area after 2D digital image stabilization compensates global motion vectors, we propose a video stabilizing algorithm which can smooth feature trajectories quickly without using homography model to estimate global motion vectors. First,we extract images＇ local feature points with improved SURF algorithm. Then,by connecting the matching points be- tween frames based on consistency of spatial motion,we get the feature trajectories. At last,we establish an objective function which takes both smoothness of the feature trajectories and degeneration of video quality into consideration to smooth the feature traiectories and get the stabilized video. The results of experiment indicate that the stabilized video with this method has 30 ~/0 less blank area and can satisfy human sense compared with those with Matsushita method, and can reduce the motion inpainting. More- over, it eliminates the accumulated error between frames in motion estimation and it is helpful to the vid- eo with large local motion in foreground.

关 键 词：电子稳像 点特征轨迹 运动估计 运动补偿 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]