二进制鲁棒不变尺度特征匹配电子稳像  

Feature matching electronic image stabilization based on binary robust invariant scalable keypionts

作　　者：崔少辉[1] 谢征[1] 王刚[1] 陈鹏[1] 

机构地区：[1]军械工程学院导弹工程系,河北石家庄050003

出　　处：《光学精密工程》2015年第9期2715-2723,共9页Optics and Precision Engineering

基　　金：军内科技创新计划资助项目(装司字[2012]665)

摘　　要：针对捷联图像制导系统中视频序列受弹体姿态变化和抖动而出现的不稳定现象,提出了一种基于二进制鲁棒不变尺度特征关键点(BRISK)的高精度快速鲁棒电子稳像算法。首先,用BRISK算法提取图像BRISK特征描述符;为了保证匹配精度和速度,采用引导互匹配策略实现BRISK特征点跟踪匹配;然后,利用改进的随机抽样一致方法(RANSAC)剔除误匹配点对;最后,利用最小二乘算法(LSA)估计全局运动参数进而实现稳像。对标准图片的匹配测试和实拍视频稳像的结果表明:结合BRISK算法的电子稳像技术,运行时间小于30ms,定位精度达到0.1pixel,对光照变化、噪声以及复杂环境遮挡具有较强的鲁棒性,能快速有效地补偿捷联制导图像的复杂随机抖动并提高视频质量。To overcome the instability of video sequences caused by attitude changes and jitters of a projectile in the strapdown image guidance system, a robust high-precision fast electronic image stabilization algorithm based on Binary Robust Invariant Salable Keypoints（BRISK） was proposed. Firstly,the BRISK algorithm was used to extract BRISK feature descriptors of the image. To improve the matching accuracy and speeds, the mutual matching strategy was adopted to realize BRISK matching feature point tracking. Then, the improved Random Sample Consensus （RANSAC） method was used to eliminate the false matching points. Finally, the Least Square Algorithm （LSA） was taken to estimate the global motion parameters to realize image stabilization. The standard image matching test and real video shoot image results show that the strapdown image guidance stabilization technology based on BRISK algorithm can provide short processing time less than 30 ms and high positioning accuracy of 0. 1 pixel. Moreover, the technology has stronger robustness to changed illumination, noise and complex environment barrier and can quickly and effectively compensate complex random jitters and improve the quality of video images.

关 键 词：捷联图像制导 电子稳像 运动估计 二进制鲁棒不变尺度特征 特征匹配 随机抽样一致方法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]