基于四参数仿射模型的频域运动估计技术  

Four Parameter Model Based Motion Estimation in Frequency Domain

作　　者：王栋[1] 王伟[1] 许晓鸣[1] 

机构地区：[1]上海交通大学电信学院智能控制实验室,上海200030

出　　处：《电子学报》2003年第1期131-134,共4页Acta Electronica Sinica

摘　　要：仿射模型能比块平移模型更好地拟合真实运动场景 ,但待解的参数也相应地增加了 .基于梯度的搜索算法在高阶问题中存在局部极小和收敛速度的缺陷 ,基于Hough变换和Gabor滤波等的变换域方法同样存在运算量的问题 ,影响了仿射模型在实时运动估计中的应用 .传统的相位相关法仅利用了傅氏频谱的相位信息求解块平移而未利用频谱的幅值特性 .本文通过对仿射运动场频域特性的分析 ,将仿射模型的线性项和平移项解耦 ,结合频谱的幅值特性和相位特性在频率域内求解仿射运动参数 .实验结果表明该算法在运算量增加不大的情况下运动估计效果得到明显改善 .Although affine model has more ability to describe the motion field,it's very difficult to be used for it's high computational complexity.In this paper,a novel fast algorithm is proposed which is derived from traditional phase correlation motion estimation method.In FFT frequency domain,the phase shift indicates the translation components of affine motion and the spectrum magnitude indicates the linear components of affine motion.So we utilize this FFT spectrum characteristic to separate the affine parameters estimation to translation components and linear components.The computational complexity is decreased dramatically.For improving the robust of linear component estimation,we use an affine invariant feature to calculate the correlative points between magnitude spectrums.Simulation indicates that our method can improve the estimation result greatly with sustainable computational complexity.

关 键 词：四参数仿射模型 频率域运动估计 相位相关 幅值相关 图像序列处理 

分 类 号：TN911.73[电子电信—通信与信息系统]