基于立体视觉的双目匹配.pdf

竺竺坚!._——————————————~一堡主兰焦笙塞———————————————————一耿工=子化比义 Abstract B1nocu胁vision is allimportant branch puter vision used in many fields·Thispaper focus on thekey ofthebinocular vision—Stereo Matching·The paper introducessome basictheoreticalknowledge aSwell asthebasic modej ofbmocular vision stereo matching,summarizes some mon stereo matc}1ingalgorithm,usesanimproved local stereo matching algorithm and an global stereo matchingalgorithm based on graph cuts,and aglobal stereo matching algont№based on Bayesian ,do some expefiments on the Middlebury international standard platform,analysis the experimental p黜d with theresultsofexistingmatching algorithms,demons仃ating the effectivenessand accuracy ofthealgorithm. Local stereo matching algorithm inthepaper use gradient di肫rence value as a matchmg cost,1mproving thesmoothness ofthedisparitymap,meanⅥ枉1e introduce ’I址ey’s Biweight function tomatching cost,SO as toachieve theeffectofsuppressing noise,in thenext step,optimize thedisparity,obtaining agoodmatching results. Intheglobal stereo matching algorithm based on graph cuts used in thepaper, leature points andedge pointsareextractedfirstly,then matchfeatu鹏points觚d edge p0Ints toobtain thedisparity ofthese points astheinitiallabelofgraph cutsalgoritluIl; calculate DAISY descriptors ofeach pixel as data item oftheenergy function. calculate thesum ofgradientdifferencebetween thehorizontal dl咒ctlon and colorchange direction asthesmooth item,finally obtaindensedisparity map bYseeking theminimum energy function,achieving stereo matclling. Intheglobal stereo matching algorithm based on Bayesiantheory usedinthe paper, during the stereo matching process,Bayesian model firSt- useMSERD∞operator to extract support points,then match thesesupponpoints wlth plxeI gray value as the matching cost,fixed window as Cost Aggregation matched,alld triangulate thematched support points tocreate two work. wetakedlsparity ofthesuppor