信息通信基于计算机视觉的建筑结构位移鲁棒性测量侯文鹏（广州大学土木工程学院广东广州5100006）摘要：传统的接触式建筑结构位移监测技术具有安装过程复杂和耗时等缺点。计算机视觉相较于传统的接触式位移监测，具有高效、成本低等优点，但大多数的计算机视觉技术在运用的过程中需要在结构上安装靶点才能实现相对准确的监测，而一些大型建筑由于结构造型等原因，给靶点的安装带来一定困难，进而影响视觉技术的准确性。针对靶点安装的难题，文章在现有的研究基础上，提出一种无需安装标靶改进的稠密光流算法，结合Python-opencv软件进行算法的编译，通过对模拟位移图像进行像素位移识别，与开源DIC算法计算结果对比验证，结果表明，稠密光流算法较开源DIC算法在计算速度上提升了40%，两种算法的位移误差在0.1%以下，并具有较好的鲁棒性。因此，本文所提出的无需靶点的算法对于一些安装靶点困难的大型建筑具有重要的参考价值。关键词：计算机视觉；光流法；非接触测量；建筑结构中图分类号：TU312.2Computervision-basedmeasurementofdisplacementrobustnessofbuildingstructuresAbstract:Traditionalcontactstructuredisplacementmonitoringtechnologyhassomedrawbackssuchascomplexinstallationprocessandtime-consuming.Comparingwithtraditionalcontactdisplacementmonitoring,computervisionhastheadvantagesofhighefficiencyandlowcost.However,mostcomputervisiontechnologiesneedtoinstalltargetsonthestructureinordertoachieverelativelyaccuratemonitoringintheprocessofapplication.Somelargebuildings,duetothestructureshapeandotherreasons,bringsomedifficultiestotheinstallationoftargets,thusaffectingtheaccuracyofvisualtechnology.Tosolvetheprob-lemoftargetinstallation,basedontheexistingresearch,thispaperpresentsadenseopticalflowalgorithmwithouttargetinstal-lationimprovement,compilesthealgorithmwithPython-opencvsoftware,identifiesthepixeldisplacementofthesimulateddis-placementimage,andcomparestheresultwiththatoftheopensourceDICalgorithm.Theresultshowsthatthedenseopticalflowalgorithmhasa40%fastercalculationspeedthantheopensourceDICalgorithm,andthedisplacementerrorofthetwoal-gorithmsislessthan0.1%.Andithasgoodrobustness.Therefore,thetarget-freealgorithmproposedinthispaperhasi...
