A 201.4 GOPS real-time multi-object recognitionprocessor is presented with a three-stage pipelined architecture.Visual perception based multi-object recognition algorithm isapplied to give multiple attentions to multiple objects in the inputimage. For human-like multi-object perception, a neural perceptionengine is proposed with biologically inspired neural networksand fuzzy logic circ
KIMetal.:A201.4GOPS496mWREAL-TIMEMULTI-OBJECTRECOGNITIONPROCESSORWITHBIO-INSPIREDNEURALPERCEPTIONENGINE
33
Fig.1.Visualperceptionbasedobjectrecognitionmodel.
pipeliningef ciency.Inaddition,amulti-castingNoCispro-posedfortheintegrationofoverall21IPblocksoftheprocessor.Forlowpowerconsumption,workload-awaredynamicpowermanagementisperformedinchip-level.Asaresult,thepro-posedprocessorachieves60frame/sec496mWmulti-object
480)recognitionupto10differentobjectsforVGA
(640
sizedvideoinput.
Therestofthispaperisorganizedasfollows.SectionIIde-scribesavisualperceptionbasedmulti-objectrecognitionalgo-rithmindetail.Then,SectionIIIexplainssystemarchitectureoftheproposedprocessor.DetaileddesignsofeachbuildingblockareexplainedinSectionIV.SectionVdescribestheproposedNoCcommunicationarchitecture.ThechipimplementationandevaluationresultsfollowinSectionVI.Finally,SectionVIIsummarizesthepaper.
II.VISUALPERCEPTIONBASEDMULTI-OBJECTRECOGNITIONA.VisualPerceptionBasedObjectRecognitionModel
Fig.1showstheconceptdiagramoftheproposedvisualper-ceptionbasedmulti-objectrecognitionmodel.Thevisualper-ceptionisanextendedmechanismofthepreviousvisualatten-tion[4]tomulti-objectcases.Basedonvisualattention,itad-ditionallyselectstheseedpointsoftheobjectsandperformsseededregiongrowingtodetecttheregions-of-interest(ROIs)http://www.77cn.com.cnparedwiththepreviousattention,thevisualperceptiongivesmultipleattentionstomultipleobjectsoftheinputimagebyhighlightingROIofeachobject.Afterthevisualperception,thenextobjectrecognitiontaskssuchaskey-pointextractionanddatabasematchingareperformedwithfocusingonlyontheselectedROIs.Byprocessingonlycriticalregionsoutofthewholeimage,computationalcostoftheobjectrecog-nitionisalsoreducedinproportionaltotheareaofselectedROIs.
B.OverallAlgorithm
Fig.2showstheoverallalgorithmoftheproposedmulti-objectrecognitionprocessor.Itisdividedintothreestagesby
theroleofeachstage:visualperception,descriptorgeneration,andobjectdecision.Thisalgorithmisdevisedtorecognizearound50of cestuffsinreal-time,whichisapplicableforautonomousmobilerobot’svisionsystem.
ThevisualperceptionstageisproposedtoestimatetheROIsofobjects,aglobalfeatureoftheimage,inadvancetomainobjectrecognitionprocessing.BasedonItti’svisualattentionmodel[8],itextractsnotonlystaticfeaturessuchasintensity,color,andorientation,butalsoadynamicfeaturesuchasmotionvectorfromthedown-scaledinputimagetogeneratesaliencymap.Basedonthissaliencymap,thevisualperceptionselectstheseedpointsofobjectsandperformsseededregiongrowingtodetectROIofeachobject[9].Finally,itdeterminestheROIsformultipleobjectsinaunitof
4040pixelsizedtile,calledagrid-tile.Fortheimplementationofvisualperceptionstage,aspecialhardwareblockwithbio-inspiredneuralnetworksandfuzzylogiccircuitsisproposedtomimicoperationsofhumanvisualsystem.
Thedescriptorgenerationstageextractskey-pointsofobjectsoutoftheselectedROIgrid-tilesfromthevisualperceptionstage,andgeneratesdescriptorvectorsforthem.Tothisend,variousalgorithmicmethodssuchasKLT,Harris-cornerdetector,af netransformations,andscaleinvariantfeaturetransform(SIFT)exist[6].Inouralgorithm,theSIFTisse-lectedbecauseitisrobusttonoiseinjectionaswellasscaleandrotationvariances.Fortheimplementationofdescriptorgenerationstage,aparallelprocessorconsistingofmanypro-cessingunitsisadoptedtotackleparallelandcompleximageprocessingtasks.Tobeapplicableforvariousalgorithms,eachprocessingunitisdesignedasaprogrammabledevice.
Theobjectdecisionstagedeterminesthe nalrecognitionre-sultsbyperformingdatabasematchingforselectedregions.Itmatchesthedescriptorvectorsoutofthedescriptorgenerationstagewiththeobjectdatabaseincludingthousandsofobjectvectors.Avectormatchingistosearchtheminimumdistancevectoroutofthevectorsinthedatabasewithaninputinquiryvector.Toacceleratetheserepeatedvectormatchingoperations,