基于结构张量的变分光流计算研究-控制理论与控制工程专业论文.docxVIP

AbstractMotion image analysis is an important task of Computer Vision and ImageProcessing,andtheopticalflowcomputationisakeytechniqueformotionimageanalysis.Infact, theopticalflowcontains abundantinformationof 3Dstructurefeatures.Itcanhelpustoknowaboutmanyimportantmotioncharacteristicsofmotionbodies.Sotheopticalflowcomputationbecomes theresearchhotspottoday,andthispaper willstudyand improvetheopticalflowcomputationtechnologybasedonvariationaltheory.Variationalcalculation for opticalflow can be attributed to solve the minimumenergyfunction.Becausesolvingthediffusion-reactionequationandminimizationofenergyfunctionareconsistent,thispapersolvesthediffusion-reactionequationandavoidstheprobleminminimizationofenergyfunction.Thispaperproposesanimprovedvariationalcalculationforopticalflowbasedonstructuretensor.Thedatatermcombinestwokindsofconstancyassumptions,whicharethebrightnessconstancyassumptionandthestructuretensorconstancyassumption.Thestructuretensor constancy assumption uses localstructuralinformation andis notsensitivetoilluminationvariation.Soitcanbebetterabletodealwithimagesequencewhichilluminationvariesandcanimprovethethe accuracyofopticalflowcomputation.Toimprovealgorithmic robustness innoise,this paperusesbothglobalandlocalmethods.Toensurethedifferenabilityandconvexityofthedata term,thispaperdesigns a non-squadratic penalizer.For smoothness term,this paper designs thediffusiontermindiffusion-reactionequationdirectly,theisotropicstrategyisusedinthelowresolutionandtheanisotropicstrategyisusedinthehighresolution.Thisdesigncan ensure the differenability and convexity and strengthen the ability ofhandlingobjectsinmotionedge.Tosolvelargedisplacementerrorsinopticalflowcomputation,thispapermakescomputationinthecoarse-to-finewarpingstrategy.Finally,thispaperprovesthattheimprovedalgorithmhasadvantageinsolvingsomeproblemsthroughseveralrepresentativeimagesequences.Analyzingtheopticalflowvectorsanderrordataoftheexperiments,thealgorithminthispaperhasimprovedinthefollo