摘要
提出一种基于Harris角点特征的稳像算法。该方法用Harris算子自适应检测出最强壮的角点,开辟(2n+1)×(2n+1)特征窗进行灰度互相关初匹配;提出欧式距离之比不变准则,通过中值滤波剔除误匹配点,确保运动矢量较高的一致性;最后保留若干正确的匹配点对,分成多组解运动矢量方程,将多组解的平均值作为全局运动矢量以提高稳像精度。实验结果表明,本算法可以克服图像缩放影响,有效消除旋转和平移抖动,实现视频图像序列稳定。
A video stabilization algorithm based on Harris is presented. In this algorithm, the strongest comer points is adaptively detected with Harris operator and （2n ＋ 1 ）× （ 2n ＋ 1 ） feature windows is selected for originally mutual gray correlative matching. Continental distance ratio invariant criterion is proposed, and consistency of motion vector is improved by eliminating error matching points with median filter. 9 pairs right points is preserved ,4 pairs composing 1 group,and 5 groups is random distributed to uncoil the motion vector equations. Last, stabilization precision is improved by averaging the 5 groups＇ solutions. The experiments show that this algorithm is unacted on image zoom, can eliminate rotation,transition and stabilize the image sequence.
出处
《激光与红外》
CAS
CSCD
北大核心
2009年第5期555-557,共3页
Laser & Infrared
基金
教育部新世纪优秀人才支持计划项目(No.NCET-05-0910)资助
关键词
电子稳像
HARRIS算子
特征点匹配
运动矢量估计
electronic image stabilization
Harris operator
feature points matching
motion vector computation
