摘要
针对稳像过程中稳像精度易受光照变化、噪声、局部遮挡等因素的影响,提出了一种区域快速鲁棒性不变特征跟踪稳像算法。首先采用改进的快速鲁棒性特征(SURF)算法提取图像局部区域特征点及其描述,然后,采用动态平衡KD树(DBKD-Tree)快速搜索匹配算法,实现局部区域特征点跟踪匹配,最后利用配准的特征点对,根据均方差最小计算稳像的全局参数实现稳像。在不同光照条件、噪声环境下进行了稳像测试,加入20%的高斯噪声时均能100%地重复检测特征,达到亚像素定位精度,误匹配率低。
An area Speed up Robust Feature （SURF） tracking algorithm for video stabilization is presented to improve the precision under the condition of illumination variation, noises, and partial occlusion. The improved SURF algorithm is used to extract the local area feature points and their descriptors. Then the fast search-matching Dynamic Balance KD Tree （DBKD Tree） is used to get the matching double points. Finally, the matching double points are used to estimate the motion parameter by the principle of minimum average variance. Experiments under different illumination and noise environments and additional 20% random noise show that the invariant feature can be 100% detected repeatedly with sub-pixel precision and near zero estimation error.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2012年第2期451-458,共8页
Journal of Jilin University:Engineering and Technology Edition
基金
'973'国家发展规划项目(2009CB72400105)
'863'国家高技术研究发展计划项目(2008AA121803)
关键词
计算机应用
电子稳像
特征提取
快速鲁棒性不变特征
运动估计
DBKD树
computer application
electric image stabilization
feather extraction
speed-up robustfeather （SURF）
motion estimation
DBKD-Tree
