摘要
提出了一种基于角点匹配的电子稳像算法。分析了Harris角点检测的原理。获得角点后,根据相邻帧对应角点的坐标分布,建立当前帧与参考帧的映射关系,并运用仿射变换模型,以最小二乘解的形式获得帧间全局运动估计矢量。最后,采用Kalman滤波器对运动估计矢量作低通滤波,平滑运动参数,获得运动补偿矢量,实现视频序列的实时稳像。实验表明该算法较好地去除了视频序列的高频抖动,同时保留了摄像机的主动运动,稳像后视频序列的峰值信噪比明显提高。
An electronic image stabilization algorithm based on corner matching is presented. First, Harris corner detection algorithm is analyzed and it is used to obtain corners of video frames. Then, the mapping relationship is built between frames on the basis of location of comers, and global motion vectors are acquired adopting least-squares solution in terms of affine transformation model. Last, the Kalman filter is employed to facilitate smooth operation as low-pass filter, motion compensated vectors between frames are acquired to smooth video sequence in real-time. Experimental results show the correct vectors of camera are reserved while dithers are eliminated, the peak signal-to-noise ratio of smoothed video is higher than that of the original video.
出处
《光学与光电技术》
2009年第4期37-40,共4页
Optics & Optoelectronic Technology
基金
军队"十一五"预研(404010204)资助项目