multi view geometry (spring '08)ocw.snu.ac.kr/sites/default/files/note/2765.pdf ·...

12
1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu Lee SoEECS, Seoul National University Projective 3D Geometry Projective 3D Geometry 2 Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU Points, Planes, Lines and Quadrics in 3D 3D points: Projective transform in : Collinear Lines are mapped to lines 15 (4x4-1) DOF 0 , ) 1 , , , ( ) , , , ( 4 4 3 4 2 4 1 3 4 3 2 1 = X X X X X X X in X X X X T T P X 3 4 3 4 2 4 1 ) , , ( R in X X X X X X T X H X 4 4× = Homogeneous representation Inhomogeneous representation 3 P

Upload: others

Post on 08-Jul-2020

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

1

Projective 3D Geometry 1

2008-1

Multi View Geometry (Spring '08)

Prof. Kyoung Mu LeeSoEECS, Seoul National University

Projective 3D Geometry

Projective 3D Geometry 2

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Points, Planes, Lines and Quadrics in 3D

• 3D points:

• Projective transform in :

CollinearLines are mapped to lines15 (4x4-1) DOF

0,)1,,,(

),,,(

44

3

4

2

4

1

34321

≠⇔

=

XXX

XX

XX

inXXXX

T

T PX3

4

3

4

2

4

1 ),,( RinXX

XX

XX T

XHX 44×=′

Homogeneous representation Inhomogeneous representation

3P

Page 2: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

2

Projective 3D Geometry 3

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Planes

• 3D planes:

0ππππ 4321 =+++ ZYX

0ππππ 44332211 =+++ XXXX4

3

4

2

4

1 ,,XXZ

XXY

XXX ===

: homogeneous planeT),,,( 4321 ππππ=π

XX' H=ππ' -TH=

Transformation

0Xπ =T

TXXXX ),,,( 4321=X : homogeneous 3D point

Projective 3D Geometry 4

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

• Euclidean representation:

• Duality:

0~. =+ dXn

( )T321 π,π,π=n

( )TZYX ,,~ =X

4π=d

n/d

points ↔ planeslines ↔ lines

: plane normal

14 =X

Page 3: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

3

Projective 3D Geometry 5

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Determination of a plane by Points

• Three points on a plane satisfies

• Or, using the coplanarity constraint;

2341DX( )T123124134234 ,,,π DDDD −−=

( ) ( ) ( )( ) ( ) ( )( ) ( ) ( )( ) ( ) ( )

0det

4342414

3332313

2322212

1312111

=

⎥⎥⎥⎥

⎢⎢⎢⎢

XXXXXXXXXXXXXXXX

[ ] 0 det 321 =XXXX

124313422341 DXDXDX +− 01234124313422341 =−+− DXDXDXDX 13422341 DXDX −

0

3

2

1

=⎥⎥⎥

⎢⎢⎢

πXXX

T

T

T

π is the null space of ⎥⎥⎥

⎢⎢⎢

T

T

T

3

2

1

XXX

πX on ∀

Projective 3D Geometry 6

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Determination of a plane Points

Page 4: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

4

Projective 3D Geometry 7

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Points from Planes

• Three planes meets at a point:

• A plane can be represented by its span:

0πππ

3

2

1

=⎥⎥⎥

⎢⎢⎢

XT

T

T

X⎥⎥⎥

⎢⎢⎢

T

T

T

3

2

1

πππ

MxX =

[ ]32134 XXX=xM

0M =Tπ

⎥⎦

⎤⎢⎣

⎡=

33

then xI

pM( )Tdcba ,,,π if = ⎟

⎠⎞

⎜⎝⎛= −−−

ad

ac

ab

,, and p

M: 3dim null-space of π

is the null space of

Projective 3D Geometry 8

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Lines – Null-space and span representation

• The line joining A and B is the span of the row space of W.

• The intersecting line of two planes P and Q is the span of the row space of W*.

BA μλ +(4dof)

⎥⎦

⎤⎢⎣

⎡= T

T

BA

W

⎥⎦

⎤⎢⎣

⎡= T

T

QP

W* QP μλ ′+′

Page 5: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

5

Projective 3D Geometry 9

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Lines – Null-space and span representation

•2 2

T T∗ ∗×= =W W WW 0

Projective 3D Geometry 10

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Points, Lines and Planes

• A plane by a point X and a line W:

• A point X by a line W and a plane :

⎥⎦

⎤⎢⎣

⎡= TX

WM 0π =M

⎥⎦

⎤⎢⎣

⎡= Tπ

WM

*

0=MX

*W

π

If the dim N(M) = 2, then X is on W

If the dim N(M) = 2, then W is on π

W

X π

π

π

Page 6: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

6

Projective 3D Geometry 11

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Lines - Plücker matrix representation

• The line joining two points can be represented by the Plücker matrix

jijiij ABBAl −= TT BAABL −=

i) L is a 4x4 skew-symmetric homogeneous matrixii) L has rank 2iii) 4 DOFiv) generalization ofv) L is independent of choice A and Bvi) Transformation

24*

×= 0LW T

yxl ×=

THLHL ='

[ ] [ ]⎥⎥⎥⎥

⎢⎢⎢⎢

⎡ −

=

⎥⎥⎥⎥

⎢⎢⎢⎢

⎥⎥⎥⎥

⎢⎢⎢⎢

=

0001000000001000

1000

0001

0001

1000

TL

Ex) x-axis

Projective 3D Geometry 12

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Dual Plücker Matrix

• Dual Plücker matrix:

• Transformation:

• Correspondence:

• Join and Incidence

1' −∗−∗ = HLHL T

TT QPPQL −=*

*12

*13

*14

*23

*42

*34344223141312 :::::::::: llllllllllll =

XLπ *=

LπX =

0* =XL

: plane through point and line

: point on line

: intersection point of plane and line

: line in plane0=Lπ

[ ] 0,, 21 =πLL K : coplanar lines

Page 7: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

7

Projective 3D Geometry 13

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Lines - Plücker line coordinates representation

• Plücker line coordinates:• From det L =0

[ ]T,l,l,l,l,ll 344223141312=L 5P∈0231442133412 =++ llllll

Klein quadric constraint

0=⇔ LL KT

[ ] 0

000001000010000100001000010000100000

34

42

23

14

13

12

344223141312 =

⎥⎥⎥⎥⎥⎥⎥⎥

⎢⎢⎢⎢⎢⎢⎢⎢

⎥⎥⎥⎥⎥⎥⎥⎥

⎢⎢⎢⎢⎢⎢⎢⎢

llllll

llllll

( ) ( )B,A,A,B, ˆˆˆ ↔LL

[ ]( )LLLL ˆ|Kˆ

ˆˆˆˆˆˆB,AB,A,det 123413421423231442133412

==

+++++=T

llllllllllll

two lines are intersect iff the 4 points are coplanar

Projective 3D Geometry 14

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Lines - Plücker line coordinates representation

• Results:

( ) [ ] 0ˆˆdetˆ == B,AA,B,|LL

( ) [ ] 0ˆˆdetˆ == Q,PP,Q,|LL

( ) ( )( ) ( )( ) 0ˆ =−= BPAQBQAP| TTTTLL

( ) 0=LL|

i) Plücker internal constraint:

ii) two lines intersect or coplanar:

4 points

4 planes

2 planes and 2 points

Page 8: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

8

Projective 3D Geometry 15

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Quadrics and dual quadrics

• Quadric:

• Dual quadric:

i) Q is a 4x4 symmetric matrixii) 9 DOFiii) in general 9 points define quadriciv) det Q=0 ↔ degenerate quadricv) pole – polar vi) (plane ∩ quadric)=conicvii) transformation

⎥⎥⎥⎥

⎢⎢⎢⎢

••••••••••

=

ooo

oo

oQ

Xπ Q=QMMC T= xX:π M=

-1-TQHHQ ='

0ππ * =QT

-1* QQ =i) relation to quadric (non-degenerate)ii) transformation THHQQ **' =

π

0XX =QT

X

Projective 3D Geometry 16

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Classification of quadrics

•• D represents Q up to projective equivalence.• Signature:

UDUQ ~T= DHHQ T=Scale normalization of D

)()( DQ σσ = Independent of H

Single planeX2=0(1,0,0,0)11

Two planesX2= Y2(1,-1,0,0)0

Single lineX2+ Y2= 0(1,1,0,0)22

Cone X2+ Y2= Z2(1,1,-1,0)1

Single pointX2+ Y2+ Z2=0(1,1,1,0)33

Hyperboloid (1S)X2+ Y2= Z2+1(1,1,-1,-1)0

SphereX2+ Y2+ Z2=1(1,1,1,-1)2

No real pointsX2+ Y2+ Z2+1=0(1,1,1,1)44

RealizationEquationDiagonalSign.Rank

Page 9: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

9

Projective 3D Geometry 17

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Classification of quadrics•

Nun-ruled quadrics: projectively equivalent to sphere:

Ruled quadrics: contains straight lines

hyperboloids of one sheet

hyperboloid of two sheets

paraboloidsphere ellipsoid

Degenerate ruled quadrics:

cone two planes

Projective 3D Geometry 18

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Twisted cubics

• A conic in Π2:

• A twisted cubic inΠ3:

⎟⎟⎟

⎜⎜⎜

++++++

=⎟⎟⎟

⎜⎜⎜

⎛=

⎟⎟⎟

⎜⎜⎜

2333231

2232221

2131211

23

2

1 1A

θθθθθθ

θθ

aaaaaaaaa

xxx

⎟⎟⎟⎟⎟

⎜⎜⎜⎜⎜

++++++++++++

=

⎟⎟⎟⎟⎟

⎜⎜⎜⎜⎜

=

⎟⎟⎟⎟⎟

⎜⎜⎜⎜⎜

344

2434241

334

2333231

324

2232221

314

2131211

3

2

4

3

2

1 1

A

θθθθθθθθθθθθ

θθθ

aaaaaaaaaaaaaaaa

xxxx

Page 10: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

10

Projective 3D Geometry 19

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

The hierarchy of transformations

⎥⎦

⎤⎢⎣

⎡vTvtAProjective

15dof

Affine12dof

Similarity7dof

Euclidean6dof

Intersection and tangency

Parallellism of planes,Volume ratios, centroids,The plane at infinity π∞

The absolute conic Ω∞

Volume

⎥⎦

⎤⎢⎣

⎡10tA

T

⎥⎦

⎤⎢⎣

⎡10tR

T

s

⎥⎦

⎤⎢⎣

⎡10tR

T

Group Matrix Distortion Invariant properties

Projective 3D Geometry 20

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

The plane at infinity

• The plane at infinity:• Contains directions (points)• two planes are parallel ⇔ line of intersection in • line // line (or plane) ⇔ point of intersection in

• The plane at infinity is a fixed plane under a projective transformation H iff H is an affinity

TXXX )0,,,( 321=D

T)1,0,0,0(=∞π

∞−

∞ =

⎟⎟⎟⎟⎟

⎜⎜⎜⎜⎜

⎥⎦

⎤⎢⎣

−==′ π

1000

1t0

ππA

AH

TT

A

∞π

∞π

∞π

Page 11: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

11

Projective 3D Geometry 21

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

The absolute conic

• the absolute conic Ω∞ is defined on π∞ s.t.

• Thus, a conic with C = I3x3

• The absolute conic Ω∞ is a fixed conic under the projective transformation H iff H is a similarity

04

23

22

21 =

⎭⎬⎫++

XXXX ( ) ( )T321321 ,,I,, XXXXXX

i) Ω∞ is only fixed as a setii) Circle intersect Ω∞ in two pointsiii) Spheres intersect π∞ in Ω∞

Projective 3D Geometry 22

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

Metric properties

• Once is identified in projective 3-space, angles and relative lengths can be measured.

• orthogonal (conjugacy)

∞Ω

Invariant to projective transform

d1, d2 : directions of two lines

(Intersection points of lines on π∞)

021 =Ω∞dd T

( )( )( )2211

21cosdddd

ddTT

T

( )( )( )2211

21cosdddd

dd

∞∞

ΩΩ

Ω=

TT

T

θ

Euclidean:

Projective:

plane

normal

Page 12: Multi View Geometry (Spring '08)ocw.snu.ac.kr/sites/default/files/NOTE/2765.pdf · 2018-01-30 · 1 Projective 3D Geometry 1 2008-1 Multi View Geometry (Spring '08) Prof. Kyoung Mu

12

Projective 3D Geometry 23

Multi View Geometry (Spring '08) K. M. Lee, EECS, SNU

The absolute dual quadric

• The Absolute dual quadric:

• The absolute dual quadric Q*∞ is a fixed quadric under the

projective transformation H iff H is a similarity

⎥⎦

⎤⎢⎣

⎡=∞ 00

0I*TQ

i) 8 DOFii) plane at infinity is the null-vector of Q∞

iii) Angles between and :

( )( )2*

21*

1

2*

1cosπQππQπ

πQπ

∞∞

∞=TT

T

θ

∞π1π 2π