基于四参数仿射模型的快速运动估计算法

Fast Motion Estimation Method Based on Four Parameter Model

为减少求解二维相位相关带来的运算量负担,提出了一种快速四参数运动估计算法。首先,进行图像幅度谱对数极坐标变换,把旋转、缩放参数降维成平移参数估计;然后,使用均值灰度投影技术替代二维相位相关,将二维数据投影到一维空间,通过一维最优相关,求解水平与垂直方向上的偏移量。实验结果表明,与相位相关技术相比,由于采用降维和快速搜索策略,求取位移的算法复杂度由O(N2log2N)下降到O(log22N),256×256图像块的计算时间由原来0.266 s下降到0.062 s,缩放精度<0.05,旋转精度<0.5°。

To reducet the computational complex of phase correlation, a new fast algorithm based on four-parameter affine model is proposed. The algorithm transforms image into a uniform polar or log-polar Fourier representation, thus rotation and scaling are reduced to translations, which are estimated by gray projection algorithm instead of phase correlation The experiments show that computational complex of estimating translations drops off compared with phase correlation. the corn puting time is reduced from 0. 266 ns to 0. 062 s with image size of 256×256, and the precision in scaling is less than 0. 05, and that in rotation is less than 0. 5°.