features-based object recognition pierre moreels california institute of technology thesis defense,...

48
Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Post on 15-Jan-2016

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Features-based Object Recognition

Pierre MoreelsCalifornia Institute of Technology

Thesis defense, Sept. 24, 2007

Page 2: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

2

The recognition continuumvariab

ility

Individual objects

means of transportation

BMW logo

Categories

cars

Page 3: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Applications

Autonomousnavigation

Identification, Security.

Help Daiki find his toys !

Page 4: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

4

• Problem setup

• Features

• Coarse-to-fine algorithm

• Probabilistic model

• Experiments

• Conclusion

Outline

Page 5: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

5

The detection problem

New scene (test image)

Models fromdatabase

Find models and their pose (location, orientation…)

Page 6: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

6

Hypotheses – models + positions

New scene (test image)

Models fromdatabase

1

2

Θ = affine transformation

Page 7: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

7

Matching features

Models fromdatabase

New scene (test image)

Set of correspondences = assignment vector

Page 8: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

8

Features detection

Page 9: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

9

Image characterization by features

• Features = high information content

‘locations in the image where the signal changes two-dimensionally’ C.Schmid

• Reduce the volume of information

edge strength map

features

– [Sobel 68]– Diff of Gaussians [Crowley84]– [Harris 88]– [Foerstner94]– Entropy [Kadir&Brady01]

Page 10: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

10

Correct vs incorrect descriptors matches

Mutual Euclidean distances in appearance space of descriptors

12

34

5

6

7

8

- Pixels intensity within a patch- Steerable filters [Freeman1991]- SIFT [Lowe1999,2004]- Shape context [Belongie2002]- Spin [Johnson1999]- HOG [Dalal2005]

Page 11: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

11

Stability with respect to nuisances

Which detector / descriptor

combination is best for recognition ?

Page 12: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

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]

Page 13: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

13

Database : 100 3D objects

Page 14: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

14

Testing setup

[Moreels&Perona ICCV05, IJCV07]

Used by [Winder, CVPR07]

Page 15: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Results – viewpoint change M

ahal

anob

is d

ista

nce

No

‘bac

kgro

und’

imag

es

Page 16: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

2D vs. 3D

Ranking of detectors/descriptorscombinations are modified whenswitching from 2D to 3D objects

Page 17: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

17

Features matching algorithm

Page 18: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

18

Features assignments

models from database

New scene (test image)

. . .

Interpretation

. . .

Page 19: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

19

Coarse-to-fine strategy• We do it every day !

Search for my place : Los Angeles area – Pasadena – Loma Vista - 1351

my car

Page 20: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Coarse-to-fine example

[Fleuret & Geman 2001,2002]

Face identification in complex scenes

Coarse resolution

Intermediate resolution

Fine resolution

Page 21: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

21

• 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 step

Coarse-to-Fine detection

Page 22: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

22

Coarse data : prior knowledge

• Which objects are likely to be there, which pose are they likely to have ?

unlikelysituations

Page 23: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

23

New scene (test image)…

Models fromdatabase

4 votes

2 votes

0 vote

Model voting

Search tree (appearance space – leaves = database features)

Page 24: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

24

(x1,y1,s1,1)

(x2,y2,s2,2)

Transform predicted by this match: x = x2-x1

y = y2-y1

s = s2 / s1

= 2 - 1

Each match is represented by a dot in

the space of 2D similarities (Hough space)

x

y

s

Use of rich geometric information

[Lowe1999,2004]

Page 25: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

• 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 transform

N~

Model

Test scene

correct transformation

Page 26: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

26Output of PROSAC : pose transformation

+ set of features correspondences

Correspondence 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.

d

d

d

[Fischler 1973] [Chum&Matas 2005]

Page 27: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

27

Probabilistic model

Page 28: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

28

Generative model

Page 29: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

29

Recognition steps

Page 30: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Score of an extended hypothesis

Hypothesis:model + position

observed featuresgeometry + appearance

database of models

constant

Consistency(after PROSAC)Prior on model

and poses

Featuresassignments

Votes per model Votes per model pose bin(Hough transform)

Prior on assignments(before actual observations)

Page 31: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

ConsistencyConsistency between observations and predictions from hypothesis

model m

position of model m

Common-frame approximation : parts are conditionally independent once reference position of the object is fixed. [Lowe1999,Huttenlocher90,Moreels04]

Con

stel

latio

n m

odel

Com

mon

-fra

me

Page 32: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

32

foreground features ‘null’ assignments

geometry geometryappearance appearance

Consistency - appearance Consistency - geometry

ConsistencyConsistency between observations and predictions from hypothesis

Page 33: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Learning 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]

Page 34: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

34

Experiments

Page 35: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

An example

Model votin

g

Hough

bins

Page 36: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

36

An example

After

PROSAC

Probabilistic

scores

Page 37: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

37

Efficiency of coarse-to-fine processing

Page 38: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

38

Giuseppe Toys database – Models

61 objects, 1-2 views/object

Page 39: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Giuseppe Toys database – Test scenes

141 test scenes

Page 40: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

40

Home objects database – Models

49 objects, 1-2 views/object

Page 41: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

41

Home objects database – Test scenes

141 test scenes

Page 42: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

42

Results – Giuseppe Toys database

Lowe’99,’04

Lower false alarmrate- more systematic verification of geometry consistency- more consistent verification of geometric consistency

undetected objects: features with poor appearance distinctivenessindex to incorrect models

-

+

Page 43: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

43

Results – Home objects database

Page 44: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

44

Failure modeTest image hand-labeledbefore the experiments

Page 45: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

45

Test – Text and graphics

Page 46: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

46

Test – no texture

Page 47: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

Test – Clutter

Page 48: Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007

48

• Coarse-to-fine strategy prunes irrelevant search branches at early stages.

• Probabilistic interpretation of each step.

• Higher performance than Lowe, especially in cluttered environment.

• Front end (features) needs more work for smooth or shiny surfaces.

Conclusions