基于结构相关的变分光流计算方法

Calculus of Variational Optical Flow Based on Structure-dependent Regularization

传统的变分光流计算方法主要是基于各种特征守恒假设和平滑性假设,对图像中的噪声、边缘、阴影、光照变化等方面的处理存在很大误差,尤其是边缘和光照变化会使计算结果产生严重误差。本文提出一种改进的变分光流计算模型,数据项采用灰度守恒与梯度守恒结合约束来解决光照变化问题。引入与结构相关的自适应平滑项,通过减弱平滑项在运动边缘的作用力度,加大梯度惩罚力度,从而有效地抑制运动物体边缘处光流计算的"漂移"现象。实验证明了模型的有效性。

The traditional calculus of variational optical flow is mainly based on various characteristic constancy assumptions and smoothness assumptions.There are serious errors in dealing with image noise,edges,shadow,illumination changes,especially the edges and illumination changes.An improved calculus of variational optical flow is proposed in this paper.The data term is to solve the problem of illumination changes by the combination of gray and gradient constancy assumption.The smoothness term related with structure-dependent regularization is introduced,which can inhibit the edges of moving objects in the ＂drift＂ phenomenon effectively by weakening the efforts of smooth term in the boundaries of the moving objects and increasing the punishment of gradient.Experimental results show that this calculus is suitable and effective.