面向视频测流的亚网格变分优化光流估计算法  

Subgrid Variational Optimization Optical Flow Estimation Algorithm for Video Flow Measurement

作　　者：宗军[1,2] 蒋东进 张亚 温宇 ZONG Jun;JIANG Dongjin;ZHANG Ya;WEN Yu(Nanjing Research Institute of Hydrology and Water Conservation Automation,Ministry of Water Resources,Nanjing 210012,China;Hydrological and Water Resources Monitoring Engineering Technology Research Center,Ministry of Water Resources,Nanjing 210012,China;Jiangsu Nanshui Water Technology Company Limited,Nanjing 210012,China;Inner Mongolia Autonomous Region Water Conservancy Engineering Construction Quality and Safety Supervision Service Center,Hohhot 010020,China)

机构地区：[1]水利部南京水利水文自动化研究所,江苏南京210012 [2]水利部水文水资源监控工程技术研究中心,江苏南京210012 [3]江苏南水水务科技有限公司,江苏南京210012 [4]内蒙古自治区水利工程建设质量与安全监督服务中心,内蒙古呼和浩特010020

出　　处：《水文》2024年第1期1-8,32,共9页Journal of China Hydrology

基　　金：国家重点研发计划资助项目“江河湖库水文要素在线监测技术与装备”(2017YFC0405705)。

摘　　要：本文在变分光流模型的基础上提出基于亚网格尺度的优化算法,用于描述图像序列中的复杂流体流动并估计其二维速度场。为解决基于网格尺度下变分光流估计中亚网格小尺度结构信息缺失的问题,结合不可压缩流体物理运动规律,引入大涡模拟的思想,在变分模型数据项中将瞬时运动分解为一个大尺度运动项和一个小尺度湍流项,并采用Smagorinsky模型求解小尺度湍流项。相比于传统的Farneback稠密光流算法,改进后的亚网格尺度变分光流算法(Subgrid scale Horn-Schunck optical flow,SGS-HS)在湍流图像序列的速度场估计中结果更优。为使SGS-HS算法同样能胜任于明渠流速测量任务,在模型的正则项方面选择使用速度梯度约束,以提高在明渠流场流向相对一致的情况下SGS-HS算法在测速实验中的精度。实验结果表明,相比于传统算法,SGS-HS算法在明渠测速方面具有较好表现。In this paper,an optimization algorithm based on sub-grid scale is studied on the basis of variable optical flow model,which is used to describe complex fluid flow in image sequences and estimate its two-dimensional velocity field.In order to solve the problem of the lack of sub-grid small-scale structural information in the estimation of variational optical flow based on grid scale,combined with the physical motion law of incompressible fluid,this paper introduces the idea of large eddy simulation,de⁃composes the instantaneous motion into a large-scale motion term and a small-scale turbulence term in the data item of varia⁃tional model,and uses Smagorinsky model to solve the small-scale turbulence term.Compared with the traditional Farneback dense optical flow algorithm,the improved Subgrid scale Horn-Schunck optical flow(SGS-HS)algorithm has better results in ve⁃locity field estimation of turbulence image sequences.In order to make the SGS-HS algorithm equally competent for the open channel flow velocity measurement task,the velocity gradient constraint is used in the regularization term of the model,which is used to improve the accuracy of SGS-HS algorithm in the velocity measurement experiment when the flow direction of the open channel flow field is relatively consistent.The experimental results show that compared with the traditional algorithm,the SGS-HS algorithm has better performance in open channel flow velocity measurement.

关 键 词：光流法 亚网格 大涡模拟 湍流 明渠 

分 类 号：TP23[自动化与计算机技术—检测技术与自动化装置] P332[自动化与计算机技术—控制科学与工程]