基于帶狀區(qū)域及其鄰接圖的立體視覺(jué)匹配技術(shù).pdf

1、中國(guó)科學(xué)技術(shù)大學(xué)碩士學(xué)位論文基于帶狀區(qū)域及其鄰接圖的立體視覺(jué)匹配技術(shù)姓名：鄔昶暢申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：模式識(shí)別與智能系統(tǒng)指導(dǎo)教師：汪增福20050501AbstractStereovisionextracts3D$GenestructurebymatchingmorethanoneimagestakenfromdifferentviewpointsAlthoughmanysolutionshavebeenproposedtothisp

2、roblem，itisstilladifficulttoproducereliabledisparitymapefficientlyThisthesispresentsanwapproachtostereocorrespondenceandthemaincon—tributionofitiSasfollows：ITheimagemodelof)StripeAdjacencyGraphs(SAGisadvancedforstereovis

3、ionInputimagesaresegmentedinto8setofnonoverlappedstripeswhichdescribesmultipleslocalfeaturesThegraphdescribesinterstriperelationshipssuchasgeometryrelationship，colorconsistencyanddisparitycontinuitystripesareverytractabl

4、einstereomatchingresultingfromtheirsimpleboundaryshape2，InthematchingmodelofstripesandSAG，eachpairofstripesconstrtlctsasurfacepatch，andthematchingoftwoinputimagesproducesanetworkofsurfacepatchThemaximumsmoothsurfaceandth

5、eglobalcostofsurfaceoutputisdefinedformatchingandevaluation3。thisthesisgiveatwostagematchingalgorithm。maximumsmoothsurfacesaxefirstlyextractedthroughsurfacegrowingaglobaloptimizationmethodbasedOniterativeoptimizationofth

6、elocalcorrespondenceisimplemented，StripesaredynamicallymodifiedtoproducebetterlocalcorrespondenceThismethodisefBcientduetotheuseoffeatureIimplementedalltheimportantalgorithmsforstripebasedstereocorrespondenceincluding：fe

7、aturebasedstripesegmentation，SAGconstructingandmodifying】max—imumsmoothsurfaceextractingandmarchingbasedonit，somelocaloptimizationmethodsandgJobaloptimizationbasedOnitExperimentalresultsshowthatotlralgcrithmiSfastandeffe