欢迎来到天天文库
浏览记录
ID:33990773
大小:3.00 MB
页数:65页
时间:2019-03-03
《基于单目视频流的前方车辆检测算法-研究》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、杭州电子科技大学硕士学位论文ABSTRACTMonocularvisionbaseddetectionsystemforfront-viewvehicleisasystemwhichaimingtoalertdriversbydetectfront-viewvehiclecorrectlyandrapidly.Withthedevelopmentofmachinevisiontechnology,thedetectionsystemiswidelyusedinthefieldsofautomateddrivingandintelligenttransp
2、ortation.Majorproblemsofthissystemarefeatureextractionandhypothesisverification.Basedontheanalysisandcompareoftheresearchsituationathomeandabroad,thispaperputforwardnewalgorithmaboutfeatureextractionandhypothesisverification.Theresearchcontentsareasfollows:(1)Ashadowfeatureextra
3、ctionalgorithmbasedonshadowlinescanningandframereduceimagefilteringisproposedbythispaperagainstmissingvehicleeasilyduringthefeatureextractionphase.Usingtheshadowareaundervehicleasthepriorifeaturewhichextractedinthisstage,thealgorithminthispagerextractshadowfeaturesandzoneaboveit
4、whichmaycontainsvehiclebyfuseafewofshadowlineswhichlocateinsameshadowarea.Becauseofthecharacteroftheshadowlineitself,thisalgorithmhassomeadvantagessuchasadaptabletoenvironment,highreal-timeperformanceandfairlylowmissingrate.Butitgetshighfalsealarmrateandwasteprocessingtime.Inord
5、ertoensurethereal-timeperformanceofsystem,theresultsetisfilteredbyframe-reduceimagesothatmostoffalsealarmcouldbeeliminated.Theexperimentverifiedthatthisalgorithmcanreducethemisdetectionratesharply.(2)Toincreasethealgorithm’stheaccuracyandreal-timeperformance,thispaperproposesave
6、rificationalgorithmicframeworkwhichbasedonvotingsystem.Afterextractingthepreliminaryvehicleinformationinscenebyfeatureextractionalgorithm,systemvotesoncandidateareadetectionthroughasetofsimplealgorithmsjustlikepositionestimation,shadowareadetection,symmetrytestandhistoryinformat
7、ion.Thenthealgorithmclassifiesthosecandidatesbythevotingresult.Thevotingsystemreducesthesizeofdatasetwhichtheverificationalgorithmneedstodealandimprovesthesystem'sreal-timeability.Becauseafewofmethodarecollaboratingtovotingsystem,thesystemismorepreciseforthosevehicleswhichlocate
8、onthesideoftheimage.(3)Tosolvingtheproblemthatt
此文档下载收益归作者所有