第52卷第1期测绘学报Vol.52,No.12023年1月ActaGeodaeticaetCartographicaSinicaJanuary,2023引文格式:李加元,张永军,艾明耀,等.渐进性优化的尺度自适应Cauchy稳健估计模型及其应用[J].测绘学报,2023,52(1):61-70.DOI:10.11947/j.AGCS.2023.20210415.LIJiayuan,ZHANGYongjun,AIMingyao,etal.Scale-adaptiveCauchyrobustestimationbasedonprogressiveoptimizationanditsapplications[J].ActaGeodaeticaetCartographicaSinica,2023,52(1):61-70.DOI:10.11947/j.AGCS.2023.20210415.渐进性优化的尺度自适应Cauchy稳健估计模型及其应用李加元,张永军,艾明耀,胡庆武武汉大学遥感信息工程学院,湖北武汉430079Scale-adaptiveCauchyrobustestimationbasedonprogressiveoptimizationanditsapplicationsLIJiayuan,ZHANGYongjun,AIMingyao,HUQingwuSchoolofRemoteSensingandInformationEngineering,WuhanUniversity,Wuhan430079,ChinaAbstract:Robustestimationisabasictechnologyingeometricprocessingandsurveyadjustment.Traditionaliterativelyreweightedleastsquares(IRLS)cannothandleproblemswithhighoutlierrates(≥50%);Randomsamplingconsensus(RANSAC)typealgorithmscanonlyobtainapproximatesolutionsandaretimeconsuming.Thispaperproposesaprogressivelyoptimizedscale-adaptiveCauchyrobustestimationmodel.First,ascaleparameterisintroducedintothetypicalCauchykernelfunctiontocontrolitsrobustness.Second,theproposedmethodusesthecontrolparametertofilteroutsomeobservationswiththelargeresidualsineachiterationandreducethetrueoutlierrate.Then,a“coarsetofine”IRLSmethodisusedforoptimizationinaprogressivemanner.Intheiterativeprocess,thecontrolparameteriscontinuouslyreducedtoimprovetherobustness.Thispaperalsoappliestheproposedmodelinseveralimportanttasksofphotogrammetry,includingmismatchremoval,imageorientation,andpointcloudregistration.Extensiveexperimentsshowthattheproposedmodelisrobusttomorethan80%outlierswhenthegrosserrorsconformtoanapproximatelyuniformorrandomdistribution,andis2~3ordersofmagnitudefasterthanRANSAC.Keywords:robustestimation;grosserrors;imagematching;spaceresection;pointcloudregistrationFoundationsupport:TheNationalNaturalScienceFoundationofChina(Nos.42...
