2024年3月16日发(作者：宝骏360发动机质量怎么样)

Volume0(1981),Number0pp.1–12

Ef?cientRANSACforPoint-CloudShapeDetection

RuwenSchnabelRolandWahlReinhardKlein

?

Universit?tBonn,ComputerGraphicsGroup

Abstract

Inthisworkwepresentorithm

decomposesthepointcloudintoaconcise,

hodisbasedonrandomsamplingand

detectsplanes,spheres,cylinders,elswithsurfacescomposedofthesebasicshapesonly,

els,weanstrate

thatthealgoriposed

methodscaleswellwithrespecttothesizeoftheinputpointcloudandthenumberandsizeoftheshapeswithin

intsetswithseveralmillionsofsamplesarerobustlydecomposedwithinlessthanaminute.

ationareasincludemeasurement

ofphysicalparameters,scanregistration,surfacecompression,hybridrendering,shapeclassi?cation,meshing,

simpli?cation,approximationandreverseengineering.

CategoriesandSubjectDescriptors

(accordingtoACMCCS)

:I.4.8[ImageProcessingandComputerVision]:Scene

AnalysisShape;SurfaceFitting;I.3.5[ComputerGraphics]:ComputationalGeometryandObjectModelingCurve,

surface,solid,andobjectrepresentations

uction

Duetotheincreasingsizeandcomplexityofgeometricdata

setsthereisanever-growingdemandforconciseandmean-

allywhendealingwith

digitizedgeometry,edwithalaserscanner,no

handlesformodi?cationofthedataareavailabletotheuser

r,inor-

dertobeabletomakeuseofthedataeffectively,theraw

digitizeddatahastobeenrichedwithabstractionsandpos-

siblysemanticinformation,providingtheuserwithhigher-

chhandlescanpro-

videtheinteractionrequiredforinvolvededitingprocesses,

suchasdeleting,movingorresizingcertainpartsandhence

canmakethedatamorereadilyusableformodelingpur-

se,traditionalreverseengineeringapproaches

canprovidesomeoftheabstractionsthatweseek,butusu-

allyreverseengineeringfocuseson?ndingareconstruction

oftheunderlyinggeometryandtypicallyinvolvesquitete-

notjusti?edinasettingwhere

acompleteanddetailedreconstructionisnotrequiredatall,

orshalltakeplaceonlyaftersomebasiceditingoperations

therhand,detecting

instancesofasetofprimitivegeometricshapesinthepoint

sampleddataisameanstoquicklyderivehigherlevelsofab-

mpleinFig.1patchesofprimitiveshapes

provideacoarseapproximationofthegeometrythatcould

beusedtocompressthepoint-cloudveryeffectively.

Anotherproblemarisingwhendealingwithdigitizedgeom-

orethe

ef?ciencyofalgorithmsinferringabstractionsofthedata

isofutmostimportance,especiallyininteractivesettings.

Thus,inthispaperwefocusespeciallyon?ndinganef?-

cientalgorithmforpoint-cloudshapedetection,inorderto

kisa

highperformanceRANSAC[FB81]algorithmthatiscapa-

bletoextractavarietyofdifferenttypesofprimitiveshapes,

whileretainingsuchfavorablepropertiesoftheRANSAC

paradigmasrobustness,

heartofouralgorithmareanovel,hierarchicallystructured

samplingstrategyforcandidateshapegenerationaswellas

anovel,lazycostfunctionevaluationscheme,whichsignif-

?

e-mail:{schnabel,wahl,rk}@

c

hedbyBlackwell

更多推荐

发动机,质量,作者