DOC PREVIEW
Berkeley COMPSCI 294 - Features-based Object Recognition

This preview shows page 1-2-3-23-24-25-26-46-47-48 out of 48 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 48 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 48 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 48 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 48 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 48 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 48 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 48 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 48 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 48 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 48 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 48 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Features-based Object RecognitionSlide 2Slide 3Slide 4Slide 5Slide 6Slide 7Features detectionSlide 9Slide 10Slide 11Slide 12Slide 13Slide 14Slide 15Slide 16Features matching algorithmSlide 18Slide 19Slide 20Slide 21Slide 22Slide 23Slide 24Coarse Hough transformCorrespondence or clutter ? PROSACProbabilistic modelSlide 28Slide 29Score of an extended hypothesisSlide 31Slide 32Learning foreground & background densitiesExperimentsSlide 35Slide 36Slide 37Slide 38Slide 39Slide 40Slide 41Results – Giuseppe Toys databaseResults – Home objects databaseFailure modeTest – Text and graphicsTest – no textureTest – ClutterSlide 48Features-based Object RecognitionPierre MoreelsCalifornia Institute of TechnologyThesis defense, Sept. 24, 20072The recognition continuumvariabilityIndividual objectsmeans of transportationBMW logoCategoriescarsApplicationsAutonomousnavigationIdentification, Security.Help Daiki find his toys !4•Problem setup•Features•Coarse-to-fine algorithm•Probabilistic model•Experiments•ConclusionOutline5…The detection problemNew scene (test image)Models fromdatabaseFind models and their pose (location, orientation…)6…Hypotheses – models + positionsNew scene (test image)Models fromdatabase12Θ = affine transformation7…Matching featuresModels fromdatabaseNew scene (test image) Set of correspondences = assignment vector8Features detection9Image characterization by features•Features = high information content‘locations in the image where the signal changes two-dimensionally’ C.Schmid•Reduce the volume of informationedge strength mapfeatures–[Sobel 68]–Diff of Gaussians [Crowley84]–[Harris 88]–[Foerstner94]–Entropy [Kadir&Brady01]10Correct vs incorrect descriptors matchesMutual Euclidean distances in appearance space of descriptors12345678- Pixels intensity within a patch- Steerable filters [Freeman1991]- SIFT [Lowe1999,2004]- Shape context [Belongie2002]- Spin [Johnson1999]- HOG [Dalal2005]11Stability with respect to nuisancesWhich detector / descriptorcombination is best for recognition ?Past work on evaluation of features•Use of flat surfaces, ground truth easily established•In 3D images appearance changes more ![Schmid&Mohr00] [Mikolajczyk&Schmid 03,05,05]13Database : 100 3D objects14Testing setup [Moreels&Perona ICCV05, IJCV07]Used by [Winder, CVPR07]Results – viewpoint change Mahalanobis distanceNo ‘background’ images2D vs. 3D Ranking of detectors/descriptorscombinations are modified whenswitching from 2D to 3D objects17Features matching algorithm18Features assignmentsmodels from databaseNew scene (test image). . .Interpretation. . .19Coarse-to-fine strategy•We do it every day !Search for my place : Los Angeles area – Pasadena – Loma Vista - 1351my carCoarse-to-fine example[Fleuret & Geman 2001,2002]Face identification in complex scenesCoarse resolutionIntermediate resolutionFine resolution21•Progressively narrow down focus on correct region of hypothesis space•Reject with little computation cost irrelevant regions of search space•Use first information that is easy to obtain•Simple building blocks organized in a cascade•Probabilistic interpretation of each stepCoarse-to-Fine detection22Coarse data : prior knowledge•Which objects are likely to be there, which pose are they likely to have ? unlikelysituations23New scene (test image)…Models fromdatabase4 votes2 votes0 voteModel votingSearch tree (appearance space – leaves = database features)24(x1,y1,s1,1)(x2,y2,s2,2)Transform predicted by this match:x = x2-x1y = y2-y1s = s2 / s1 = 2 - 1Each match is represented by a dot inthe space of 2D similarities (Hough space)xysUse of rich geometric information[Lowe1999,2004]•Prediction of position of model center after transform•The space of transform parameters is discretized into ‘bins’•Coarse bins to limit boundary issues and have a low false-alarm rate for this stage•We count the number of votes collected by each bin.Coarse Hough transformN~ModelTest scenecorrect transformation26Output of PROSAC : pose transformation + set of features correspondencesCorrespondence or clutter ? PROSAC•Similar to RANSAC – robust statistic for parameter estimation•Priority to candidates with good quality of appearance match•2D affine transform : 6 parameters each sample contains 3 candidate correspondences.ddd[Fischler 1973] [Chum&Matas 2005]27Probabilistic model28Generative model29Recognition stepsScore of an extended hypothesisHypothesis:model + positionobserved featuresgeometry + appearancedatabase of modelsconstantConsistency(after PROSAC)Prior on modeland posesFeaturesassignmentsVotes per modelVotes per model pose bin(Hough transform)Prior on assignments(before actual observations)ConsistencyConsistency between observations and predictions from hypothesismodel mposition of model mCommon-frame approximation : parts are conditionally independent once reference position of the object is fixed. [Lowe1999,Huttenlocher90,Moreels04]Constellation modelCommon-frame32foreground features‘null’ assignmentsgeometrygeometryappearanceappearanceConsistency - appearance Consistency - geometryConsistencyConsistency between observations and predictions from hypothesisLearning foreground & background densities•Ground truth pairs of matches are collected•Gaussian densities, centered on the nomimal value that appearance / pose should have according to H•Learning background densities is easy: match to random images.[Moreels&Perona, IJCV, 2007]34ExperimentsAn exampleModel votingHoughbins36An exampleAfterPROSACProbabilisticscores37Efficiency of coarse-to-fine processing38Giuseppe Toys database – Models61 objects, 1-2 views/objectGiuseppe Toys database – Test scenes141 test scenes40Home objects database – Models49 objects, 1-2 views/object41Home objects database – Test scenes141 test scenes42Results – Giuseppe Toys databaseLowe’99,’04Lower false alarmrate- more systematic verification of geometry consistency- more consistent verification of geometric consistencyundetected objects: features with poor appearance distinctivenessindex to incorrect models-+43Results – Home objects database44Failure modeTest image hand-labeledbefore the experiments45Test – Text and graphics46Test – no textureTest – Clutter48•Coarse-to-fine strategy prunes irrelevant search branches at early stages.•Probabilistic interpretation


View Full Document

Berkeley COMPSCI 294 - Features-based Object Recognition

Documents in this Course
"Woo" MAC

"Woo" MAC

11 pages

Pangaea

Pangaea

14 pages

Load more
Download Features-based Object Recognition
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 Features-based Object Recognition 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 Features-based Object Recognition 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?