Improvement of optical flow obtained by compensation method using multilayer resolution structure
Improvement of optical flow obtained by compensation method using multilayer resolution structure
复制标题
使用多层分辨率结构的补偿方法获得的光流的改进
DOI：
10.1109/iccas.2016.7832350
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Hiroshi Harada
中科院分区：
文献类型：
--
作者：
Ryota Hirakawa;Teruo Yamaguchi;Hiroshi Harada
光流场是物体像的速度分布，我们采用时空微分的方法得到光流场。该方法可以快速获得光流场，但由于速度太快，可能会产生较大的系统误差。因此，我们引入了用于时空微分的减小速度幅值的补偿方法。在补偿中，我们从物体的真实的速度中减去补偿速度，然后我们得到的差值作为扰动速度。如果我们能适当地设置补偿速度，扰动速度接近于零，我们可以估计它的误差很小。为了确定补偿速度，我们使用多层构造图像。然而，如果我们降低图像的分辨率，则可以获得光流的区域变得更小。在这项研究中，我们试图科普这个问题。此外，我们试图提高光流估计的准确性。
Optical flow is referred to as velocity distribution of object image, and we have used spatio-temporal differentiation method to obtain optical flow. It is possible to obtain optical flow fast by the method, but the high velocity may induce large systematic errors. So, we introduced the compensation method for reducing the magnitude of velocity which is applied to spatio-temporal differentiation. In the compensation, we subtract compensation velocity from the real velocity of the object, and then we obtain the difference as a perturbation velocity. If we can set the compensation velocity appropriately, the perturbation velocity approaches to zero, and we can estimate it with little error. To determine a compensation velocity, we use the multilayer structural image. However, if we lower the resolution of the image, the region in which optical flow may be obtained becomes smaller. In this study we try to cope with this problem. In addition, we try to improve accuracy of optical flow estimate.