单目多视角显微三维重建关键技术分析

《单目多视角显微三维重建关键技术分析》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、浙江工业大学硕士学位论文MULTI-VIEW3DRECONSTRUCTIONUSINGMONOCULARMICROSCOPICIMAGESABSTRACTMicroscopic3Dreconstruction,asacoretechniqueofmicro-visionsystem,hasaveryimportantapplicationvalueinprecisionmanufacturing.However,becauseofobviousmicroscopicimagedegradation,3Dreconstructionfrommulti-viewimagesneedsto

2、beimprovedfrommanyaspectswhenitisappliedinmicrofield.Thegoalofthispaperistoestablishasimpleandefficientmicroscopic3Dreconstructionsystemwithmonocularmicroscopyandapplyittothesurfacereconstructionofmicro-scaleobjects,thusprovidingessentialreferencesformicroscopicinspection.Themaintasksandachieve

3、mentsareasfollows:1.Comparativestudyofimagefocusingevaluationfunction.Imagefocusingevaluationfunctionisoneoftheimportantbasistomakemicro-visionsystemfocusrapidly.Ninespecificimagefocusingevaluationfunctionsofthreekindshavebeenstudied.Howtherobustnessofdifferentkindsofevaluationfunctionchangewhe

4、nmanificationandnoiseintensitychangewasanalyzedthroughexperiments.Thefocusingqualityevaluationschemeofmicroscopicimagecollectionwasfinallydeterminedwiththestudyresults.2.Thestudyofthekeypointdetectionandmatchingalgorithmofimages.Keypointdetectionandmatchingarecrucialelementsofreconstructionprec

5、ision.Thischapterintroducedconstraintbasedonlocalgeometricconsistencyoftransformonthebaseofperformancecomparisonoffourusualkeypointdetectionalgorithmsandthencreatedanewoutlierremovalalgorithm.Thisalgorithmcanobtainmorematchingpointsandmeanwhilereducethematchingerrorsubstantially.Also,itcanelimi

6、natematchingmistakesnearbyepipolarlinewhichcannotbeinspectedbyRANSACandthelike.ii万方数据浙江工业大学硕士学位论文3.Developasetofroundedmicroscopic3Dreconstructionsystem.Themicroscopic3Dreconstructionsystemhardwareinthispaperconsistsof4-DOFmotionplatformandmicroscopicimagingsystem,soitcancollectmulti-viewmicros

7、copicimageseasily.Thenucleusmoduleofthesystemsoftwareintegratedmicroscopicimagecapturingmethod,microscopicimagekeypointdetectionandmatchingalgorithm,andfurthercombinedStructureFromMotionframetoachieveefficientmicroscopic3Dreconstr