考虑遮挡情况的光流场估计算法

Occlusion Reasoning Based Optical Flow Estimation

遮挡是影响光流场估计的主要因素.针对其破坏光流场估计的区域颜色一致性假设,从而导致现有的光流场估计算法失效的问题,我们提出一种基于遮挡推理的光流场估计算法.首先构建一个图像金字塔,然后在每层结构中以形变补偿图像(warped image)为纽带,探索其中"重影"区域和遮挡区域的内在关系,并结合邻域关系等信息获得遮挡推理线索,最后将之融入优化函数中通过迭代优化的形式解决存在遮挡时的光流场估计问题.我们在Flying Chairs和MPI-Sintel等数据集上进行了相关实验验证了本文算法的有效性.

Occlusion is the main factor affecting the optical flow estimation. It can destroy the assumption of color consistency,and causes the existing optical flow estimation algorithm to fail. To handle such situations,we propose an occlusion reasoning based optical flow estimation algorithm. Firstly,a coarse-to-fine image pyramid structure is constructed. Then,the warped image is used as a link to explore the intrinsic relationship between the ghosting artifacts and the occlusion regions,and the occlusion reasoning clues are obtained by combining the information of neighborhood relationship. Finally,it is integrated into the optimization function to estimate the optical flow in the presence of occlusion through iterative optimization. Experiments on Flying Chairs and MPI-Sintel datasets demonstrate the effectiveness of our method.