一种用于电子稳像的运动矢量处理方法

Motion vector processing method for electronic image stabilization

在典型的电子稳像系统中,电子稳像包括运动估计和运动补偿两部分。运动估计得到的运动矢量包含了相机的有意运动和无意运动。在运动补偿中,我们直接利用运动估计得到的运动矢量平移像素,经常造成过补偿现象。针对这种情况,提出了一种基于一阶线性自回归的运动矢量处理方法(运动平滑),对运动矢量进行处理,分离出相机的无意运动(抖动)矢量,然后进行运动补偿,实验结果表明,该算法在没有降低系统运算速度的前提下得到了较好的稳像效果。

In a typical electronic image stabilization system, tow parts, i.e. motion estimation and motion compensation, are included. The motion vectors obtained from motion estimation are composed of tow kinds of camera movements： intended movement and unmeant movement. In the motion compensation, the motion vector that obtained from motion estimation to move pixels is directly used, and it often leads to the phenomenon of excessive compensation. In view of this situation, a processing method is proposed in the paper, which is based on first-order linear autoregressive to process the motion vector that is from excessive compensation （here is called motion smooth ）. Then the vector came from camera movement is shortlisted, which belongs to unmeant movement （here is called jitter）. The experiment results show that this algorithm gives a better result under the premise that it does not reduce the operation speed of whole system.