基于双分数阶变分光流模型的运动目标检测方法

变分光流法是常用的运动目标检测方法,应用场景中的光照变化会极大影响现存变分光流法的稳定性及准确率,提出基于分数阶的变分光流模型来提高光照变化鲁棒性.该模型将分数阶导数同时应用于经典变分光流模型的数据项及平滑项中;根据图像的数据特征,采用图像函数及光流向量函数的有限离散二重分数阶导数近似拟合求导结果计算光流向量,进而将光照变化的影响按比例分摊到了各周边像素点中,弱化了光流向量计算过程中光照变化对光流模型的影响.为避免同一像素点的二重分数阶导数覆盖了不同光流向量区域的问题,采用超像素分割来舍弃不在同一光流向量区域的像素点,从而保证了同一光流向量区域的整体性,从而进一步提高模型的光流估计准确率.在三大常用数据库及户外视频序列中进行实验,实验证明,在应用场景存在光照变化的前提下,基于双分数阶的变分光流模型优于其他算法;对比其他算法,文中算法的运算速度适中.

A Motion Detection Method for Dual Fractional-order Variational Optical Flow Model

Motion detection is a basic problem in the field of image processing and optical flow model is one of the frequently used motion detection method.To enhance the robustness of optical flow estimation in illumination changes scene,a dual fractional-order variational optical flow model is proposed.The model is a fractional-order version of HS(Horn&Schunck)model.In particular,the fractional-order derivative is applied in the brightness and smoothing constraint equation of HS model.Two dimensional fractional differential masks are proposed and applied to process the simplification of the model.In order to solve the isotropy of dual fractional differential mask,super-pixel segmentation is applied to adjust the shape of the mask,so as to limit the mask into one super-pixel region with a same or similar motion style.Using fractional-order derivative in smoothness constraint equation makes the model preserving the discontinuity of the optical flow model.Using two-dimensional fractional differential mask,the brightness value and optical flow value of pixels surround target points would be merged into target points,do so the robustness of the model to illumination changes improves.What’s more,two-dimensional fractional differential mask is also applied to calculate dual fractional fractional-order variational optical flow model,do so calculate procedures are simplified and the algorithm is easily programmed.Experiments are done in three widely used public dataset and image sequences extracted from outdoor videos,experiment results show that compared to conventional variational optical flow model,the superiority of our model is the robustness in illumination changes scenes;Compared with other algorithms,the operation speed of the algorithm in this paper is moderate.