DOC PREVIEW
MSU CSE 803 - HANDOUTS

This preview shows page 1-2-3-27-28-29 out of 29 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 29 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

1Segmentation, Classification &Tracking of Humans for Smart Airbag Applications Dr. Michael E. FarmerDept. of Computer Science, Engineering Science, and PhysicsUniversity of Michigan-FlintImportance of Topic• Between 1986 and 2001 airbags caused the following deaths, mostly during low speed crashes:– 19 infants in RFIS – 85 children – 7 adults due to proximity at time of deployment• Unfortunately, as recently as 1998:– over 50% of a NHTSA survey respondents had placed their children in the front seat of their vehicle over the last 30 days• Deaths may continue until a system to automatically disable the airbag is developed.2Automotive Occupant Classification &Tracking Problem• National Highway Transportation Safety Administration (NHTSA) mandate (2004)• Requires two suppression conditions:•Suppress the airbag if child or infant•Detect change within 10 seconds•Suppress the airbag if occupant is within Automatic Suppression Zone (ASZ) •Disable bag within 20 msecsDefinition of Automatic Suppression Zone (ASZ)• Oriented along Instrument Panel– Distance from Airbag where risk of injury in minimizedASZASZ3NHTSA Approved Methods for Performing Airbag Suppression1-year old infant 3 to 6-year old childAdult Suppress ifPresentSuppress ifPresentSuppress ifin ASZSuppress ifin ASZProtected byClassificationProtected byTrackingOR•If decide to protect children by classification then get 4-class problem (infant, child, adult, empty)•If decide to protect children by tracking then get 2-class problem (infant versus all else)4-class Classification ProblemInfant ChildAdult Empty42-class Classification ProblemVs.InfantsAdultsLarge Intra-class VariabilityDifficulties with Occupant ClassificationAdult ClassInfant Class5Difficulties with Occupant Classification II6 Year-old on Booster 5th % Adult FemaleNote that the 6 YO is actually taller in these imageswhen seat is in forward position for child and rear-mostposition for the adult.Low Inter-class Variability for 4-Class Classification Problem- System experiences extreme variations in illuminationDifficulties with Occupant Classification6Difficulties with Occupant TrackingExtensive occupant deformation during movementOccupant occlusionVariability in size of occupantSummary of Difficulties with Occupant Airbag Suppression Problem• Large intra-class variability of the various occupant types• Low inter-class variability for 4-class classification problem• Camouflaged classes (e.g., blanketed RFIS)• Large variation in illumination• Severe automotive environmental conditions•Low cost• Extremely high reliability and performance7Summary of Image Data UsedOccupant Type Classification Number of ImagesRFIS+FFIS Infant 2657Child Child 620Adult Adult 983Empty Seat Empty 72Total number of images:4332Occupant Type Classification Number of ImagesInfant (RFIS+FFIS) Infant 1807Child Child 236Adult Adult 210Empty Seat Empty 8Total number of images:2261Training images:Test images:System Algorithm ArchitectureTracker SegmenterOccupant ModelHead/TorsoTracker&PredictorInputImageClassifier SegmenterFeatureExtractionOccupantClassifierEvery 3 secondsEvery 1/40 second• Standard components of pattern recognition system:1. data acquisition and pre-processing2. data representation3. decision making8Approach for Classifier SegmentationRawImageClassifier SegmenterFeatureExtractionOccupantClassifier• Pal and Pal state: “hundreds of segmentation techniques in the literature, but there is no single method which can be considered good for all images”• Pal and Pal also state:“semantics and prior information about the type of images are critical to the solution of the segmentation problem”• In light of these, we will utilize all the information we have regarding the interior of the vehicle• The approach we will take is one of background removal. Segmentation Processing• For Background Removal tested 2 methods– De-correlation Processing–Eigen-images– Results:• De-correlation processing outperforms Eigen-images• Also developed 2 post-processing methods– Hole filling using binary morphology• Use closing: – Further background reduction with Watershed()BBA Θ⊕ˆ9Eigen-image Processing)(newnewTMbnewµ−= IΦpnewnewMbbackgroundµ+= pΦIthresholdbackgroundnew || >−IITΦCΦL =MbΦUsed sequence of722 images tocompute covariancematrixDe-correlation Processing[]Tyxgradgradji ,),( =g∑∑⋅∑∑⋅=ABAByxyxyxyxC22),(),(),(),(gggg10Eigen-image Processing ResultsInput Image TransformedImageDifferenceImageHistogram ofDifference ImageThresholdedDifference ImageSegmentation Processing ResultsInput ImageReference ImageThresholdedDe-correl. ImageDe-correlationImageInput ImageHole FillingWatershed ImagePost-processing11Approach for Feature ExtractionRawImageClassifier SegmenterFeatureExtractionOccupantClassifier• Devijer and Kittler define Feature Extraction as: – “extracting from the raw data the information which is most relevant for classification”. – “feature extraction is probably the single most important factor in achieving high recognition performance”Feature Extraction TaxonomyContent RetrievalMethodsTextureDescriptionColorDescriptionTransformDomainSpatial(Geometric)DomainStructuralSpatialLocationTransformDomainSpatial(Geometric)DomainRegion-basedMethodsShapeDescriptionBoundaryMethods12Feature Extraction- Child on seat has same boundary as empty seat so cannot use boundary methods- Child and infant have common color/grayscale & texturedistribution so cannot use color/grayscale or textureRecall wesegment the occupant &the seat- We will use region methods to characterize the occupant Legendre Moments of Edge Image for Feature ExtractionSegmentedImagesEdgeImages13Legendre Moment ReconstructionAdult ImageInfant Image25 Order(351 features)35 Order(666 features)45 Order(1081 features)Will use thehighest orderto capture themost internalfeaturesResults for Feature Extraction• Tested the following different moment features:– Geometric, Legendre, and Zernike• Recall we do not want moment invariants, since size of occupant depends on spatial locationTested moments based on separation of features using Mann-Whitney distribution test14Feature Selection TaxonomyWill test our own Mann-Whitney as well as Mutual Information for Best FirstFeature SelectionMethodsRandomHeuristicNon-ExhaustiveForwardSelectionBackwardSelectionAll BreadthFirstBranch&BoundBestFirstBeamSearchInstanceBased(RELIEF)Type I Type


View Full Document

MSU CSE 803 - HANDOUTS

Documents in this Course
Load more
Download HANDOUTS
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view HANDOUTS and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view HANDOUTS 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?