一种用于图象帧间主运动估计的快速线搜索方法

A RAPID LINE SEARCH METHOD USED IN GLOBAL MOTION ESTIMATION BETWEEN CONSECUTIVE VIDEO FRAMES

本文中作者提出了一种新的基于鲁棒统计的快速线搜索方法 ,可以用于图象帧间主运动估计 ,能够提高算法速度 .近年来 ,一种新的参数估计技术—鲁棒统计—被越来越广泛地用于主运动估计 ,与传统的基于最小二乘的估计方法相比较 ,鲁棒统计对于外点具有更好的鲁棒性 ,但运算复杂度较高 .而主运动估计中耗时最大的部分是线搜索 ,因此我们针对鲁棒统计中常用的 M估计函数形式 ,采用近似函数拟合的方法 ,设计了一种快速的线搜索方法 ,与牛顿法结合 ,可以很大提高主运动估计算法速度 .

A new rapid line search method based on robust statistics is presented in this paper, which could be used in global motion estimation between consecutive video frames for speeding up objective. In recent years, robust statistics based methods are widely used in parameter estimation. Compared with the classical least mean square method, robust statistic based method is more robust to outliers, but is also more complex for calculation. Because the line search step occupies most of the computation cost for global motion estimation, we developed an approximate fitting function based line search method according to the M estimation function of robust statistics. Combined with Newton method, our method gets great gains in speed.