Device and method for detecting object and device and method for ...

2 downloads 303 Views 2MB Size Report
Nov 22, 2004 - Primary Examiner * Daniel Mariam. Issued? Aug- 11' 2009. (74) Attorney, Agent, or Firm * Frommer Lawrence
USO0RE43 873E

(19) United States (12) Reissued Patent

(10) Patent Number:

Hidai et a]. (54)

US RE43,873 E

(45) Date of Reissued Patent:

DEVICE AND METHOD FOR DETECTING

Dec. 25, 2012

OTHER PUBLICATIONS

OBJECT AND DEVICE AND METHOD FOR

X-

iangrongi

Ch

en e

t 1, “L a . I

-

t t-

earning represen a lVe

1

0C3.

1 f m ea

res

f

or

face detection” Proceedings 2001 IEEE Conference on Computer _

.

.

.

.

_

Vision and Pattern Recognition. CVPR 2001. Kauai, Hawaii, Dec.

(75) Inventors gaegelclrltloggéabgé?ongpl%25:21:12‘; Toky’o (JP) -



_

8-14, 2001, Proceedings of the IEEE Computer Conference on Com ’

puter Vision and Pattern Recognition, Los Alamitos, CA, IEEE Comp. Soc, US, vol. vol. 1 of 2, Dec. 8, 2001 (Dec. 8, 2001), pp.

.

1126-1131, XP010583872 ISBN: 0-7695-1272-0.

(73)

Asslgnee' Sony corporatlon’ Tokyo (JP)

(21)

Appl' NO" 13/208’123

ColmenareZ A J et al: “Face detection with information-based maxi

_ .

(22) R .

mum discrimination” Proceedings. 1997 IEEE Computer Society

Conference on Computer Vision and Pattern Recognition (Cat. No.

_

97CB36082) IEEE Comput. Soc Los Alamitos, CA, USA, Jun. 1997

Flledf

Aug- 11’ 2011

(Jun. 1997), pp. 782-787, XP002312941 ISBN: 0-8186-7822-4.

Related US. Patent Documents

(Continued)

e1ssue 0 :

(64)

Patent No.: Issued?

7,574,037 Aug- 11’ 2009

Primary Examiner * Daniel Mariam (74) Attorney, Agent, or Firm * Frommer Lawrence &

APP1- NOJ

10/994,942

Haug LLP; William S. Frommer

Filed:

Nov. 22, 2004

(57) (

30

)

F

'

Al't'

P"tDt

orelgn PP lea Ion non y

Nov. 25 2003

a a

(JP) ............................... .. 2003-394556

(51) Int_ CL G06]; 9/62

gradation image~ A Scaling Section generates Scaled images

by sealing down a gradation image input from an image output section. A scanning section sequentially manipulates the scaled images and cutting out window images from them

(200601) (200601)

G06]; 9/00

ABSTRACT

An object detecting device for detecting an object in a given

and a discriminator judges if each window image is an object

(52)

us. Cl. ...................................... .. 382/159' 382/103

Or 110‘- The discn'minator includes a Plurality of Weak dis

(58)

Field of Classi?cation Search

382/103

criminators that are learned in a group by boosting and an

218 224’

adder for making a weighted majority decision from the out puts of the weak discriminators. Each of the weak discrimi

382/118

See application ?le for comp’lete gearcil hist’ory ’ '

(56)

nators outputs an estimate of the likelihood of a window

References Cited

image to be an object or not by using the difference of the luminance values between two pixels. The discriminator sus

pends the operation of computing estimates for a window image that is judged to be a non-object, using a threshold

US. PATENT DOCUMENTS 6,711,279 B1

3/2004 HamZa et al.

7,050,607 B2 * 7,054,489 B2 *

5/2006 5/2006

2002/0102024 A1

value that is learned in advance.

Li et al. ....................... .. 382/118 Yamaoka et al. ........... .. 382/203

72 Claims, 18 Drawing Sheets

8/2002 Jones et a1.

1

image output section

\L Scaling sec?on

1 Scanning section

1

Referring to

Gruup learning machine

Discriminator

~2

Next scaled image

US RE43,873 E Page 2 OTHER PUBLICATIONS

Comp. Soc, US, V01. V01. 1 of 2, Dec. 8, 2001, pp. 1126-1131,

Marcel S et al: “Biometric face authentication using pixel-based

XP010583872 ISBN: 0-7695-1271-0. ColmenareZ A J et al: “Face detection With information-based maxi

Weak classi?ers” Biometric Authentication. ECCV 2004 Interna

tional Workshop, BioaW 2004. Proceedings (Lecture Notes in

Comput. Sci. vol. 3087) Springer-Verlag Berlin, Germany, May 2004 (May 2004), pp. 24-31, XP002312942 ISBN: 3-540-22499-8. Xiangrong Chen et al: “Learning representative local features for face detection” Proceedings 2001 IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2001. Kauai, Hawaii, Dec. 8-14, 2001, Proceedings of the IEEE Computer Conference on Com

puter Vision and Pattern Recognition, Los Alamitos, CA, IEEE

mum discrimination” Proceedings. 1997 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CAT. No. 97CB36082) IEEE Comput. Soc Los Alamitos, CA, USA, Jun. 1997, pp. 782-787, XP002312941 ISBN: 0-8186-7822-4. Marcel S et a1: “Biometric face authentication using pixel-based Weak classi?ers” Biometric Authentication. ECCV 2004 Interna

tional Workshop, BioaW 2004. Proceedings (Lecture Notes in

Comput. Sci. vol. 3087) Springer-Verlag Berlin, Germany, May 2004, pp. 24-31, XP002312942 ISBN: 3-540-22499-8.

* cited by examiner

US. Patent

Dec. 25, 2012

Sheet 1 0f 18

US RE43,873 E

US. Patent

Dec. 25, 2012

Sheet 2 0f 18

US RE43,873 E

1548

,1 548-2

RELATED ART

US. Patent

Dec. 25, 2012

E

Sheet 4 0f 18

E

N:

w: 3

w:

US RE43,873 E

.wdE

US. Patent

Dec. 25, 2012

Sheet 5 0f 18

US RE43,873 E

“E@982“ N);

c$3om52a:?5"0

v) a

2

5.3m

a

52 8 2“58m50g

mGE

US. Patent

Dec. 25, 2012

Sheet 6 0f 18

US RE43,873 E

wecu5mowr?iucmwoE

0. -“

US. Patent

Dec. 25, 2012

Sheet 7 0f 18

10A

FIG.7

US RE43,873 E

US. Patent

5

Dec. 25, 2012

Sheet 8 0f 18

US RE43,873 E

xEzn3ig;ma

g is

“a528r%e.“;?

1LaBE3W>w,Mu mm

iEN»E;.

(s2%;

wdE

US. Patent

Dec. 25, 2012

Sheet 9 0f 18

US RE43,873 E

Luminance value 11

321 Luminance value I z

Inter-pixel difference characteristic = I 1 - I2

US. Patent

Dec. 25, 2012

Sheet 10 0f 18

US RE43,873 E

Threshold value Th

yi =\_1

—>

>

FIG ° 1 0A

g

1i

a2

l/

I)

LL-

‘yr’

,

,1’

,_

\\

Y1 = 1

\\\

/\\

Inter-pixel difference characteristic Formula 3

FlG.1 OB

FIG.1OC

vi=1

A. V

Formula 5 ———————————— —-

Distribution of non-object data

( yi : '1)

Distribution of object data ( vi = 1)

US. Patent

Dec. 25, 2012

FIG.11A

Sheet 11 0f 18

US RE43,873 E

3a.zwc2o"v. inter-pixel difference characteristic --—-------—--- Distribution of non-object data

( yi : "1)

Distribution of object data ( vi =

FIG.11B Inter-pixel difference characteristic

US. Patent

Dec. 25, 2012

Sheet 12 0f 18

US RE43,873 E

TUEN

US. Patent

Dec. 25, 2012

(

Sheet 13 0f 18

Start learning

US RE43,873 E

)

Initialize data weight of ieaminz sample

Select a weak discriminator

~82

l Compute discrimination error

of weak discriminator

Compute weight (reliability of weak discriminator) of weighted majority decision

l Update data weight of learning sample

l Compute detection suspension threshold value

N / o

Predatermined number of times K of learning session (predetermined number K of weak discriminators)

\

over? Yes End

FIG.13

ivS6

87

US. Patent

Dec. 25, 2012

C

Sheet 14 0f 18

Start learning of weak discriminator >

US RE43,873 E

)

i

Randomly select two pixels l, ,l, M 81 1

Determine interpixel difference characteristic for all learning samples and compute frequency distribution

~3|2

I

Compute threshold value T“, of weak discriminator for producing minimum value e mi. of weighted error

~$13

l

Compute threshold value Tm, of weak discriminator for producing minimum value cm of weighted error

~$14

l

Determine pixel positions and threshold value for them

emu(em':S‘ em

Suggest Documents