【摘要】 从光流的基本约束方程出发，提出了一种基于区域平滑约束的光流场估计方法。首先将图像按其运动特征划分成不同的区域，然后对各区域中的光流场施加各向非均匀的有条件平滑约束，将光流场的估计转化为求解一大型线性方程组，最后用神经网络模型将算法予以实现。实现结果表明，该算法准确稳定，适用于多物体有遮掩的场景。更多还原

【Abstract】 This paper proposes an optical flow estimation method based on the flow constraint equation. First the image is segmented into several regions according to the motion features, and then the optical flow is estimated in each region under a kind of weighted anisotropic smoothness constraint. Finally a recurrent neurual network is proposed to achieve a possible real time solution. This new approach leads to the accurate and robust results especially for multi object motion.更多还原