相互结构引导滤波TV-L1变分光流估计

由于光流场既包含物体的运动信息，又包含场景的三维结构信息，因此光流计算技术是计算机视觉和机器视觉领域研究的重要任务之一.针对现有光流计算方法在图像边缘保护方面存在过度平滑问题，提出一种基于相互结构引导滤波的TV-L1（Total Variational with L1 norm，TV-L1）变分光流估计方法.通过提取置信度较高的图像相互结构区域，构造基于相互结构引导滤波的全局目标函数，并采用金字塔分层细化与交替迭代方案结合的策略进行优化，该方法可以较好的保护图像边缘信息.最后采用标准测试图像集对本文方法与现有代表性变分方法LDOF（Large Displacement Optical Flow，LDOF），CLG-TV（Combined Local-Global Total Variation，CLG-TV），Classic++，NNF（Nearest Neighbor Fields，NNF）以及深度学习方法FlowNet2.0进行对比，实验结果表明本文方法具有较高的光流估计精度与鲁棒性，尤其对图像边缘保护具有显著的效果，并且在运动目标检测，机器人避障等方面具有一定应用前景.

Mutual-Structure Guided Filtering Based TV-L1 Optical Flow Estimation

Because the optical flow field contains both the motion information of the object and the three-dimensional structure information of the scene,optical flow calculation technology is one of the important tasks in the field of computer vision and machine vision.But,for the existing optical flow methods,there is over-smoothing problem in image boundary preserving.This paper proposes a global TV-L1(Total Variational with L1 norm,TV-L1) variational optical flow computation method based on the mutual-structure guided filtering.By extracting the mutual-structural regions of the image with higher confidence,we construct the global mutual-structure guided filtering objective function,and optimize the algorithm via combining the pyramid layering strategy with the alternating iteration scheme.This method can better preserve the image boundary information.Finally,we compare the proposed method with the existing representative variational methods LDOF(Large Displacement Optical Flow,LDOF),CLG-TV(Combined Local-Global Total Variation,CLG-TV),Classic++,NNF(Nearest Neighbor Fields,NNF) and deep learning method FlowNet2.0 by using standard test image datasets.The experimental results demonstrate that the presented method has more accuracy and better robustness than the other evaluated methods,especially has the significant effect of boundary preserving in the areas,and it has application prospects in moving target detection,robot obstacle avoidance,and so on.