基于范数优化路径的视频稳像算法  

Video stabilization algorithm based on norm optimal paths

作　　者：石春雷[1,2] 杨文佳[2] 黄培康 

机构地区：[1]光学辐射重点实验室,北京100854 [2]北京环境特性研究所,北京100854 [3]电磁散射重点实验室,北京100854

出　　处：《系统工程与电子技术》2015年第6期1411-1416,共6页Systems Engineering and Electronics

摘　　要：针对运动相机中视频的随机抖动和低频晃动问题,提出了一种基于范数优化路径的视频稳像算法。该算法在特征点粗匹配之后引入概率进化抽样一致性算法剔除误匹配点,避免了局部运动对全局运动估计的影响,然后在定制的L1范数约束下构建相机的最优化运动路径,最后利用新的路径进行运动矫正获取稳定的视频图像序列,实验结果验证了算法的有效性。To deal with the problem of random shakiness and low frequency bounces, an improved video stabilization method based on norm optimal camera paths is proposed. Probability evolution sample consensus algorithm is used to eliminate the error matching points after coarse matching of feature points, avoiding the in terference of local motion. Then camera paths are gained by L1 norm optimization. Finally, image warp is com puted to get stabilized video sequences. Experimental results for different video sequences show the proposed method has a better veracity of stabilized results.

关 键 词：视频稳像 概率进化抽样 L1范数优化 相机运动路径 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]