资源描述:
《[2000 CVPR] A Statistical Method for 3D Object Detection Applied to Faces and Cars》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、AStatisticalMethodfor3DObjectDetectionAppliedtoFacesandCarsHenrySchneidermanandTakeoKanadeRoboticsInstituteCarnegieMellonUniversityPittsburgh,PA15213Abstractisspecializedtofrontalviews.Weapplytheseview-basedInthispaper,wedescribeastatisticalmethodfor3Dobjectde
2、tectorsinparallelandthencombinetheirresults.Iftherearedetection.Werepresentthestatisticsofbothobjectappear-multipledetectionsatthesameoradjacentlocations,ouranceand“non-object”appearanceusingaproductofhisto-methodchoosesthestrongestdetection.grams.Eachhistogra
3、mrepresentsthejointstatisticsofasubsetWeempiricallydeterminedthenumberoforientationstoofwaveletcoefficientsandtheirpositionontheobject.Ourmodelforeachobject.Forfacesweusetwoview-baseddetec-approachistousemanysuchhistogramsrepresentingawidetors:frontalandrightpr
4、ofile,asshownbelow.Todetectleft-varietyofvisualattributes.Usingthismethod,wehavedevel-profilefaces,weapplytherightprofiledetectortoamirror-opedthefirstalgorithmthatcanreliablydetecthumanfacesreversedinputimages.Forcarsweuseeightdetectorsaswithout-of-planerotationa
5、ndthefirstalgorithmthatcanreli-shownbelow.Again,wedetectleftsideviewsbyrunningtheablydetectpassengercarsoverawiderangeofviewpoints.sevenright-sidedetectorsonmirrorreversedimages.1.IntroductionThemainchallengeinobjectdetectionistheamountofvariationinvisualappear
6、ance.Forexample,carsvaryinshape,size,coloring,andinsmalldetailssuchasthehead-lights,grill,andtires.VisualappearancealsodependsontheFigure1.Examplesoftrainingimagesforeachfaceorientationsurroundingenvironment.Lightsourceswillvaryintheirintensity,color,andlocati
7、onwithrespecttotheobject.Nearbyobjectsmaycastshadowsontheobjectorreflectadditionallightontheobject.Theappearanceoftheobjectalsodependsonitspose;thatis,itspositionandorientationwithrespecttothecamera.Forexample,asideviewofahumanfacewilllookmuchdifferentthanafron
8、talview.AnobjectdetectormuchaccommodateallthisvariationandstilldistinguishtheobjectfromanyotherpatternthatmayoccurinthevisualFigure2.Examplesoftrainingimagesforeachcarorientationwo