结合运动矢量和像素递归的全局运动估计方法

Global motion estimation method with motion vectors and pixel recursion

全局运动估计是计算机视觉、视频处理等领域中一项重要研究课题。结合运动矢量和像素递归提出一种新的全局运动估计方法,该方法根据块运动矢量求出运动矢量直方图,找出主要块运动方向作为初始的全局运动方向,并初始化全局运动参数。利用运动矢量间距离及类间方差求出运动矢量分割阈值,自适应地去除外点块区域。根据背景块梯度和值的大小,在每个背景块中选择一到两个特征像素点进行运动参数估计。实验结果表明,该方法具有较快的计算速度,同时也具有较高的计算精度。

Global motion estimation is one of the important research topics in video processing and computer vision. In this paper, a new approach for global motion estimation is proposed by combining the block motion vectors and pixel reeursive algorithm. The proposed method first establishes a vector histogram based on the block motion vectors, and then finds the main block motion vector as initial global motion orientation, to set the initial motion parameters. The threshold of motion segmentation is obtained and outlier areas are adaptively eliminated by using the distance between motion vectors and the variance between clusters. One or two feature pixels in each background block are selected to estimate the global motion parameters according the gradient sum. Experimental results show that the proposed algorithm is accurate, fast, and efficient.