一种结构向量与分数阶相结合的光流模型

An Fractional Order Optical Flow Model Combined with Structure Tensor

运动目标检测是图像处理中的最基本任务,变分光流法是常用的运动目标检测技术。因模型固有的缺陷,传统变分光流模型不能应对场景中的光照变化,无法保留运动不连续性,从而局限了变分光流模型的应用范围,影响了光流估计的计算精度。提出一种局部结构不变约束与分数阶平滑约束相结合的光流模型,该模型应用局部结构张量函数代替传统亮度约束方程中的亮度函数,应用分数阶导数代替传统平滑约束方程中的整数阶梯度。因结构张量是一个不随光照变化的物理量,且包含了图像中的一些结构信息,而分数阶平滑约束能保留边缘不连续性,因而使得模型能在各种光照环境下获得清晰的运动目标轮廓。实验证明所提模型对光照变化鲁棒,且能保留光流场边缘不连续性,较其他相关模型,能获得更高准确率的光流场。

In order to obtain clear optical flow motion object outline under challenging illumination changes,an improved fractional order optical flow model was proposed.Specifically,the data term of the model proposed is composed by structure tensor constancy assumption,and the smoothness term of our model is make up by fractional order smoothness constraint.For the ability of preserving the discontinuity of different motion regions of fractional order smoothness constraint and the illumination invariant of the structure tensor,our model can get clear motion outline and obtain high accurate optical flow estimation.Experiment results show that our model is robust to illumination changes and can preserve the discontinuity of different motion regions.