What is computer vision ?

munchsistersΤεχνίτη Νοημοσύνη και Ρομποτική

17 Οκτ 2013 (πριν από 4 χρόνια και 22 μέρες)

122 εμφανίσεις

Slide1
Whatiscomputervision?
Computervisionisconcernedwithmodelingandreplicatinghuman
visionusingcomputersoftwareandhardware.Itcombines
knowledgeincomputerscience,electricalengineering,mathematics,
physiology,biology,andcognitivescience.Itneedsknowledgefrom
allthesefieldsinordertounderstandandsimulatetheoperationof
thehumanvisionsystem.
Slide2
Whatiscomputervision?(cont’d)
Computervision(imageunderstanding)isadisciplinethatstudies
howtoreconstruct,interpretandunderstanda3Dscenefromits
2Dimagesintermsofthepropertiesofthestructurespresentin
thescene.
Slide3
Typicalhardwarecomponentsofacomputervisionsystem
IR light
Computer
frame grabber
controller
Micro
Alarm
Camera
Wefocusoncomputervisionalgorithmsandtheirsoftware
implementation.
Slide4
ComputerVisionHierarchy
•Low-levelvision:processimageforfeatureextraction(edge,
corner,oropticalflow).
•Intermediate-levelvision:objectrecognitionand3Dscene
interpretationusingfeaturesobtainedfromthelow-levelvision.
•High-levelvision:interpretationoftheevolvinginformation
providedbytheintermediatelevelvisionaswellasdirecting
whatintermediateandlowlevelvisiontasksshouldbe
performed.Interpretationmayincludeconceptualdescription
ofascenelikeactivity,intentionandbehavior.
Slide5
ComputerVisionHierarchy(cont’d)
Slide6
WhyIsComputerVisionDifficult?
•Theproblemisill-posedinverseproblem.
•Noisyimagedataordatawithuncertainties.
Slide7
RelatedFields
Computervisionoverlapssignificantlywiththefollowingfields:
imageprocessing,patternrecognition,andphotogrammetry.
Imageprocessingfocusesonimagemanipulationtoenhanceimage
quality,torestoreanimageortocompress/decompressanimage.
Mostcomputervisionalgorithmsusuallyassumesasignificant
amountofimageprocessinghastakenplacetoimproveimage
quality.
Patternrecognitionstudiesvarioustechniques(suchasstatistical
techniques,neuralnetwork,supportvectormachine,etc..)to
recognize/classifydifferentpatterns.Patternrecognitiontechniques
arewidelyusedincomputervision.
Photogrammetryisconcernedwithobtainingaccurateandreliable
measurementsfromimages.Itfocusesonaccuratemensuration.
Cameracalibrationand3Dreconstructionaretwoareasofinterest
Slide8
tobothcomputervisionandphotogrammetryresearchers.
Slide9
ComputerVisionv.s.ImageProcessing
Imageprocessingstudiesimage-to-imagetransformation.The
inputandoutputofimageprocessingarebothimages.Typical
imageprocessingoperationsinclude
•imagecompression
•imagerestoration
•imageenhancement
Slide10
ComputerVisionv.s.ImageProcessing(cont’d)
Computervisionistheconstructionofexplicit,meaningful
descriptionsofphysicalobjectsfromtheirimages.Theoutputof
computervisionareadescriptionoraninterpretationorsome
quantitativemeasurementsofthestructuresinthe3Dscene.Image
processingandpatternrecognitionareamongmanytechniques
computervisionemploystoachieveitsgoals.
Slide11
ExampleApplications
•Robotics
•Medicine
•Security
•Transportation
•Industrialautomation
•Image/videodatabases
•HumanComputerInterface
Slide12
RoboticsApplications
•Localization-determinerobotlocationautomatically
•Obstaclesavoidance
•Navigationandvisualservoing
•Assembly(peg-in-hole,welding,painting)
•Manipulation(e.g.PUMArobotmanipulator)
•HumanRobotInteraction(HRI):Intelligentroboticsto
interactwithandservepeople
Slide13
Figure1:NASAroverforforplanetarysurfaceexploration
Slide14
Figure2:Avision-guidedweldingmachine
Slide15
Figure3:Realtimevisualservoingforrobotgrasping
Slide16
Figure4:HRI:companionrobot
Slide17
IndustrialAutomation
•Industrialinspection(defectdetectionandmensuration)
•Assembly
•Barcodeandpackagelabelreading
•Objectsorting
•Documentunderstanding(e.g.OCR)
Slide18
GeometricTolerancing
215
(a)Originalleftimage(b)Originalrightimage
(a)Detectedcornersandholeboundaries(b)Detectedcornersandholeboundaries
1
2
3
4
(c)Reconstructed3Dholeboundaries
Figure11.3:Resultsoffeatureextractionand3Dreconstructionforpart3
Slide19
Medicine
•Classificationanddetection(e.g.lesionorcellsclassification
andtumordetection)
•2D/3Dsegmentation
•3Dhumanorganreconstruction(MRIorultrasound)
•Vision-guidedroboticssurgery
Slide20
MedicalImaging
Slide21
MedicalImaging(cont’d)
Slide22
Security
•Biometrics(iris,fingerprint,facerecognition)
•Surveillance-detectingcertainsuspiciousactivitiesorbehaviors
Slide23
Security:FaceDetectionandRecognition
Slide24
Slide25
Transportation
•Autonomousvehicle
•Safety,e.g.,drivervigilancemonitoring
Slide26
Image/VideoDatabaseSearch/Retrieval
Itismainlyusedforimageretrievalbasedonimagecontent.
Slide27
Slide28
HumanComputerInterface
•Gazeestimation
•Faceexpressionrecognition
•Headandhandgesturerecognition
Slide29
HeadPoseandGaze
Slide30
ComputerVisionLiterature
1.Journals
•IEEEtransactionsonPatternRecognitionandMachine
Intelligence(PAMI)
•InternationalJournalofComputerVision
•Computervisionandimageunderstanding
•Imagevisionandcomputing
•Machinevisionandapplication
•Patternrecognition
2.Conferences
•Internationalconferenceoncomputervision(ICCV)
•IEEEconferenceoncomputervisionandpattern
recognition(CVPR)
•Internationalconferenceonimageprocessing(ICIP)
Slide31
•Internationalconferenceonpatternrecognition(ICPR)
•IEEEconferenceonroboticsandautomation
Slide32
ComputerVisionResources
ComputerVisionInformationPageshttp://www.visionbib.com
•Publications
•Visiongroups
•Software
•Conferences
•Imagedatabases
•Vendorsandcompanies
Additionallinksforcomputervisionmaybefound
http://www.cns.nyu.edu/∼eero/vision-links.html
http://www.cs.berkeley.edu/∼daf/book.html
Slide33
ImageProcessingResources
•computervisionnewsgroup:
http://www.vislist.com/
•imageprocessingnewsgroup:
sci.image.processing
•FundamentalsofImageProcessing
http://www.ph.tn.tudelft.nl/Courses/FIP/noframes/fip.html
•AnImageProcessingTutorial
http://www.cs.washington.edu/research/metip/tutor/tutor.html
Slide34
Topics
•ImageAcquisitionandFormation
•PerspectiveProjectionGeometry
•CameraCalibrationandPoseEstimation
•3DReconstruction
3Dreconstructionasingleimage(shapefromX)
3Dreconstructionfrommultipleimages
•MotionEstimationandTracking
Opticalflowestimation
ObjecttrackingwithKalmanfiltering
Structurefrommotion
•FeatureExtraction(Edge,point,line,curve)
Slide35
BackgroundNeeded
•Patternrecognitionandmachinelearning
•Numericalanalysis
•Statistics
•Linearandnon-linearoptimizationandregression
•Programmingskills
•Computationalgeometry
•Projectivegeometry
•Digitalsignalprocessing
•Physics
Slide36
Outcomes
•understandthefundamentalcomputervisiontheories
•havetheabilitytodesignandimplementcertaincomputer
visiontechniques
•havethecapabilityofapplyingcomputervisiontechnologiesto
applicationsofinterest.
Slide37
ReferencestoComputerVisionTerminologies
1.DictionaryofComputerVisionandImageProcessing,Robert
Fisher,KenDawson-Howe,AndrewFitzgibbon,CraigRobertson,
EmanueleTrucco,Wiley,2005.
2.R.M.HaralickandL.G.Shapiro,“GlossaryofComputer
VisionTerms,”PatternRecognition24:69-93,1991.
3.R.M.Haralick,“GlossaryandindextoRemotelySensedImage
PatternRecognitionConcepts,”PatternRecognition5:391-403,
1973.