视频分析中全局运动矢量快速估计的研究

Research on rapid estimation of global motion vector of video

全局运动一般是由于摄像机移动或转动引起,但通常摄像过程中发生转动的概率很小,所以在全局运动估计中可以忽略旋转参数,运用简单的四参数全局运动模型。迭代最小二乘估计法求出运动模型中的4个参数,再根据解出的运动方程实现全局运动估计。为提高算法实时性,改变传统上对图像中所有块进行遍历的方法,而仅对图像中的特殊子集块进行迭代计算,使得运算复杂度大幅下降。实验显示在保证不降低精度的前提下,估计速度提3高倍左右。

Global motion is generally caused by camera movement or rotation, but the probability for camera rotating is very low. So, the rotation parameters can be neglected during the global estimation, using simple four parameter of global motion model. The iterativelestsquares estimation method is used to find out the four parameters of the motion model and then the global mo tion estimation is achieved according to the motion equation. In order to improve the realtime, in stead of adopting traditional search full blocks in the image, iterative estimation can be taken on a special a subset of the block to make the computation less complex. The experiments show that the computing speed will be improved about three times when the precision is guaranteed.