一种用于实时数字稳像的全分辨率运动补偿方法

以室外视频监控中帧间平移运动为研究对象,提出一种基于平衡位置修正与自适应存储帧更新的实时全分辨率数字稳像方法。首先利用卡尔曼滤波在线估计镜头的主动运动,预先确定抖动图像补偿后的平衡位置;然后针对图像补偿后产生的空白区域,通过选择合适的参考帧及预存帧构成存储帧队列,根据存储帧队列的覆盖范围对平衡位置进行修正,实现修正后空白区域像素的全部拼接填补;最后以修正后的平衡位置为基准,根据同时满足左上、右上、左下和右下4个偏移方向的原则对参考帧队列进行自适应在线更新,用最新采集帧替换时间上较早、空间内容上冗余的参考帧,使得存储帧队列在有限长度内保证尽可能大的空间覆盖。本文提出的方法可以满足实时数字全分辨率稳像的要求,适用于集成式、嵌入式等各类视频监控系统。 

A full-resolution motion compensation method used for real-time digital image stabilization

With the foundation of inter-frame translation motion of the videos captured in outdoor video monitoring, a real-time video stabilization method was presented with equilibrium position modification and adaptive update of storage frames, which can realize the full-resolution video stabilization. Firstly, the Kalman filter was used to estimate the intentional motion of the camera, in which the equilibrium position of the currently stabilized frame was obtained. Secondly, the reference frames and subsequent frames were stored. Based on the spatial coverage of the storage frames, the equilibrium position was modified to reach the full-resolution in-painting of the compensated current frame. Finally, taking the modified equilibrium position as the standard, and guaranteeing the coexistence of the storage frames with deviation in upper left, upper right, lower left, and lower right directions, we substituted the newest captured frames online for the earliest and redundant reference frames, so the storage frames can achieve the greatest coverage and temporal-spatial correlation with limited storage resource. The presented video stabilization algorithm achieves the full-resolution stabilization and real-time processing, which can be applied to the integrated and embedded video surveillance systems.