Saliency-based Discriminant Tracking(2009CVPR)

《Saliency-based Discriminant Tracking(2009CVPR)》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Saliency-basedDiscriminantTrackingVijayMahadevanNunoVasconcelosDepartmentofElectricalandComputerEngineeringUniversityofCalifornia,SanDiegovmahadev@ucsd.edu,nuno@ece.ucsd.eduAbstractingboxatvideoframet,aclassifieristrainedtodistinguishtargetfeaturesfromthoseofthebackground.Thisclassi-

2、Weproposeabiologicallyinspiredframeworkforvisualfieristhenusedtodeterminethelocationofthetargetintrackingbasedondiscriminantcentersurroundsaliency.framet+1.Theboundingboxismovedtothislocation,Ateachframe,discriminationofthetargetfromtheback-theclassifierupdated,andtheprocessiterated.g

3、roundisposedasabinaryclassificationproblem.FromaInthebiologicalworld,objecttrackingiscloselyrelatedpooloffeaturedescriptorsforthetargetandbackground,atoattentionaltasks,suchastheguidanceofeyemovements.subsetthatismostinformativeforclassificationbetweentheDuetotheevolutionaryadvantages

4、ofsolvingthesetaskstwoisselectedusingtheprincipleofmaximummarginaldi-accurately,itisnotsurprisingthatbiologicalvisionsystemsversity.Usingthesefeatures,thelocationofthetargetinthehavedevelopedextremelyefficienttrackingmechanisms,nextframeisidentifiedusingtop-downsaliency,complet-interm

5、sofbothaccuracyandspeed.Theeffectivenessofingoneiterationofthetrackingalgorithm.Wealsoshowthesemechanisms,evenunderthemostadverseconditionsthatasimpleextensionoftheframeworktoincludemotion(e.g.highlyclutteredscenes,low-light,etc),isaconse-featuresinabottom-upsaliencymodecanrobustlyi

6、den-quenceoftheavailabilityofrobustsaliencymechanisms,tifysalientmovingobjectsandautomaticallyinitializethethatcausepre-attentivepop-outofsalientlocationsinthetracker.Theconnectionsoftheproposedmethodtoexistingvisualfield[23].Thesesalientlocationsbecomethefocusworksondiscriminanttrac

7、kingarediscussed.Experimen-ofattention(FoA)forthepost-attentivestagesofvisualpro-talresultscomparingtheproposedmethodtothestateofcessing,wheretop-downfeedbackfromhigherlevelcorti-theartintrackingarepresented,showingimprovedperfor-callayersisusedtosolveproblemssuchastrackingorvi-manc

8、e.sualsearch[31]wit