中文摘要：

提出一种图像全景补偿输出的实时稳像系统。首先,提出了基于子区域运动迭代的全局运动估计算法,来快速有效地求取图像平面的仿射运动参数:在求取各子区域运动时,采用灰度投影相关的三点自适应搜索策略,结合大小搜索步长,进行由粗到细的搜索定位;然后将各子区域运动带入6参数仿射运动模型,通过最小二乘迭代,去除误差较大块即误匹配或前景运动块,从而提高全局运动估计精度。其次,对原始运动矢量序列进行滤波处理,以提取抖动参数:采用Kalman滤波器,一方面较好地平滑运动矢量以降低视频抖动,另一方面有效跟随摄像系统的有意扫描运动。最后,提出图像全景补偿方法,有效改善图像边界信息缺失带来的视觉误差。实验结果表明该系统可以实时处理视频序列的平移旋转等抖动,而且输出全景视频图像保持视觉的完整流畅。

英文摘要：

A real time implementation is presented to stabilize video sequence with panoramic images as output.Firstly,the gray projection algorithm based on the least-square iteration is proposed to estimate six-parameter affine matrix.In order to reduce the computation cost of the full correlation,the three points’ adaptive search method combing large and small search steps is presented to realize motion estimation from-coarse-to-fine.And then the motions of divided areas are iteratively computed with the least-squa...