2基于联合驱动的变分光流计算模型-中国图象图形学报.doc

中图法分类号：TP391.41 文献标识码：A 文章编号： - 论文索引信息： 图像光流联合驱动的变分光流计算新方法( 梅广辉1),陈震1),危水根1)，张聪炫2) 1) (南昌航空大学计算机视觉研究室，江西 南昌 330063) 2) (南京航空航天大学自动化学院，江苏 南京 210016) 摘 要：本文提出一种基于图像光流联合驱动的变分光流计算方法。数据项采用灰度守恒和梯度守恒相结合、局部约束与全局约束结合的思想，并引入正则化因子提高计算精度。平滑项采用图像与光流联合驱动的各向异性平滑策略，将数据项与平滑项紧密地联系起来，并通过设计扩散张量的两个本征值来控制光流扩散速度。最后采用多分辨率分层细化策略解决大位移问题。实验证明，该计算模型在背景复杂、光照变化、运动边界等情况的光流计算具有很好的效果。 关键词：光流；变分法；正则化；图像光流驱动；扩散 New algorithm for estimation of variational optical flow with Image-and Flow-driven MEI Guanghui1) , CHEN Zhen1) , WEI Shuigen1) ,Zhang Congxuan2) 1) (Laboratory of Computer Vision, Nanchang Hangkong University, Nanchang,Jiangxi, China 330063) 2) (College of Automation Engineering, Nanjing University of Aeronautics and Astronautics,Nanjing,Jiangsu,China 210016) Abstract: In this paper, a computation technology of variational optical flow based on Image-and Flow-driven is proposed. The thought to obtain the data term is to use the combination of gray and gradient constancy assumption, as well as the combination of local and global constraint, and the regularization factor is introduced to improve the accuracy. A joint image- and flow-drive anisotropic smoothing strategy is introduced to design the smooth term, linking the data term and smooth term closely. And then, the speed of optical flow can be controled by the design of the eigenvalues of diffusion tensor. Finally, the paper utilizes the multiscale hierarchical refinement strategy to handle large displacements problem. The experiment results proved that the model had a good computation effect of optical flow to cope with the problem of additive illumination changes and the boundaries of the moving objects with complicated background. Key words: Optical flow;Variational method;Regularization;Image-and Flow-driven;Diffusion0 引言 光流计算在视觉运动的研究中有非常重要的作用。所谓光流是指图像中灰度模式表观运动的速度，它不仅包含了被观察物体的运动信息，还携带了丰富的3D结构信息。光流计算被广泛地应用于图像分割、识别、跟踪、机器人导航、运动估计和3D结构重建等计算机视觉与图像处理领域。 光流计算技术的研究开始于上世纪80年代初Horn和Schunck [1]以及Lucas和Kanade[2]等人奠基性的工作，而后的20年里各种新的算法和改进算法不