针对装配机器人金属高光件特征提取技术研究
Abstract
Theaccurateacquisitionofgeometricfeaturesandparametersofmechanicalpartsisthe
premiseandkeyoftheassemblyrobottocompletetheassemblytask.Timelyandaccurate
acquisitionofthegeometricfeaturesandparametersofthetargetpartscanimproveproduction
efficiencyandensureproductqualityandproductionsafety.Metalpartsarewidelyusedin
actualproductionandassemblywork,andtheirsurfaceshavespecialreflectivepropertiesto
light.Duetothecomplexfieldenvironmentanddifferentlightconditions,localorevenlarge
areaofhighlightreflectionisoftenproducedwhenindustrialcamerasareusedforimage
acquisitionofmetalparts,whichmakesthefeatureinformationofmetaltargetscoveredbythe
highlight,resultinginthelossoftargetdataandreducingtheoverallqualityoftheimage.In
thispaper,thefollowingresearchiscarriedoutonthesituationthathighlightsappearinthe
imageofmetalpartsinthehighlightenvironmentintheassemblyoperationoftheassembly
robot,whichcoversthegeometricfeaturesofthetargetandleadstolowdetectionrate:
1)Inordertonarrowthesearchscopeoftargetextractionandreducetheinfluenceof
complexbackgroundfactors,theYOLOv5targetdetectionalgorithmisproposedtobeapplied
totheroughlocationofpartstargets.TheexperimentalresultsshowthatthemAPvalueofthe
proposedmodelisupto96.6%forgasketrecognitionand95.9%forgeardetection,which
provestheaccuracyoftheproposedmodelandmeetstheneedsofactualtargetdetection.
2)Aimingattheproblemthathighlightsappearintheimageofmetalpartsduringimage
acquisition,coveringthegeometricfeaturesofthetargetandresultinginlowdetectionrate,a
highlightextractionandrecoveryalgorithmbasedonthecombinationofsaliencydetectionand
fastmovingrepairisproposed.