简介:Conventionalcameracalibrationthatemployscalibrationtargetsisacommonlyusedmethodtoacquireacamera'sintrinsicand/orextrinsicparameters.Thecalibrationtargetsareusuallydesignedasperiodicarraysofsimplehigh-contrastpatternsthatprovidehighlyaccurateworldcoordinatesystempointsandthecorrespondingimagepixelcoordinatesystempoints.Theexistingpixelcoordinateextractionalgorithmscanreachasub-pixellevel;however,theytreateachsinglepatterninoneimageasanindependentindividual,whichmakesitdifficulttofurtherimproveextractionaccuracy.Inthispaper,anovelmethodisproposedbyutilizingtheperiodicarrangementcharacteristicsofthecalibrationtargetpatternasaglobalconstrainttoimprovethecalibrationaccuracy.Basedonacamera'spinholemodel,theintersectionpointoftwofittedcurvesisusedasanoptimizedpixelpointtoreplacetheinitialone.Followingthepixelcoordinateoptimizationprocedures,experimentswereperformedusingrealdatafroma3Dlaserlinescannerandadynamicprecisioncalibrationtarget.Ourresultsshowthattherelativeerrorsofcamerahomographymatrixelementsobtainedbytheproposedoptimizationmethodwerereducedcomparedwiththecommonlyusedmethod.Theaveragecoordinatemeasurementaccuracycanbeimprovedbynearly0.05mm.Itisshownthattheproposedoptimizationmethodcanenhancethecameracalibrationaccuracy,especiallywhentheextractedpixelsareofpoorerprecision.