基于帧间特征点匹配的流速测量模型研究  

Research on flow velocity measurement model based on inter-frame feature point matching

作　　者：刘明堂[1] 张浩[1] LIU Mingtang;ZHANG Hao(Faculty of Electronic Engineering,North China University of Water Resources and Hydropower,Zhengzhou 450045,China)

机构地区：[1]华北水利水电大学电子工程学院,河南郑州450045

出　　处：《水利水电快报》2025年第3期116-123,共8页Express Water Resources & Hydropower Information

基　　金：河南省重大科技专项项目(201110210300);郑州市重大科技创新专项项目(2019CXZX0050);河南省高等学校重点科研项目计划(21A510007)。

摘　　要：图像法在流速测量领域越发重要,提出了基于帧间特征点匹配的流速测量模型,在试验模型上进行了验证,并在黄河濮阳范县段进行了应用。对加速稳健特征(Speeded Up Robust Features,SURF)算法与暴力匹配(Brute Force,BF)算法进行改进,根据图像中的轮廓信息对河道区域进行分割,使得SURF算法的识别区域与识别结果更精准,识别速度更快。然后,根据特征点间的欧氏距离筛选特征点,计算出特征点的像素位移,根据帧速率计算得到特征点的像素速度并将其转化为实际速度。结果表明:与多普勒流速仪测量结果进行对比,所提算法测量结果的平均绝对误差为0.56 cm/s,平均相对误差为5.7%;与LK光流法相比,分别提高了0.3 cm/s与4.1%。该算法应用于黄河濮阳范县段的分析结果表明:平均相对误差为4.24%,比LK光流法提高了4.15%。基于帧间特征点匹配的流速测量模型在保证测量精度的情况下,能兼顾流速测量的实时性与便捷性,满足对河流流速实时、准确、便捷的测量要求。The image method plays an increasingly important role in the field of flow velocity measurement.We proposed a flow velocity measurement algorithm based on inter frame feature point matching,which had been validated on experimental models and applied in the Fanxian section of Yellow River.We also improved SURF(Speeded Up Robust Features)and BF(Brute Force)by segmenting the river area based on the contour information in the image,making SURF′s recognition area and results more accurate and faster.Then,feature points were filtered based on the Euclidean distance between them,and the pixel displacement of the feature points was calculated.The pixel velocity of the feature points was calculated based on the frame rate and converted into the actual velocity.The result showed that the average absolute error of the algorithm measurement results in this paper was 0.56 cm/s,and the average relative error is 5.7%compared with the measurement results of the Doppler current meter.Compared with the LK optical flow method,it had increased by 0.3 cm/s and 4.1%,respectively.The algorithm was applied to the Fanxian section of Yellow River in Puyang,with an average relative error of 4.24%,which was 4.15%higher than the optical flow method.The flow velocity measurement algorithm based on inter frame feature point matching can balance the real-time and convenience of flow velocity measurement while ensuring measurement accuracy.It can basically meet the real-time,accurate and convenient measurement requirements for river flow velocity in the context of water conservancy digitization and informatization.

关 键 词：流速测量 智能检测 图像法 帧间特征点图像处理 

分 类 号：TB937[一般工业技术—计量学] TN911.73[机械工程—测试计量技术及仪器]