针对现有的光流方法在处理大位移和估计运动图像细节方面存在的问题,提出一种结合图像细节特征的变分光流场模型.首先通过增加特征点的对应,采用自适应的保持边缘的正则项以及引入occlusion检测函数对经典光流模型进行了改进；其次,采用基于变分框架下的高斯金字塔方法以及加权中值滤波的方法对所提出的模型进行求解.大量的实验结果表明,文中模型有效地解决了以上问题,并取得了较好的效果.%There exist some problems in processing large displacement and estimating motion details with modern optical flow. A new variational optical flow method combining feature details of images is proposed. First, the classical optical flow model is improved by adding the correspondence of the feature points, adopting the adaptive edge-preserving regularization and introducing the occlusion detection function. Second, variational framework together with the Gaussian pyramid and weighted median filter method are applied in the solving process. Finally, abundant experiment results show that the proposed method solves above problems effectively and achieves good results.
展开▼
