Bag of lines models - Center For Machine Perception...
Transcript of Bag of lines models - Center For Machine Perception...
![Page 1: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/1.jpg)
“Bag of lines” modelsfor non-central linear cameras
![Page 2: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/2.jpg)
Guillaume BATOG
INRIA Nancy Grand-Estproject-team VEGAS
(joint work with X. Goaoc and J. Ponce)
“Bag of lines” modelsfor non-central linear cameras
![Page 3: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/3.jpg)
pinholepushbroom
pencil
linearoblique
Linear cameras
X-slit
![Page 4: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/4.jpg)
Goal
Stereo-vision betweenANY pair of
linear cameras
![Page 5: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/5.jpg)
PROBLEMS unified modelcomputations
{Goal
Stereo-vision betweenANY pair of
linear cameras
![Page 6: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/6.jpg)
PROBLEMS unified modelcomputations
{[Yu & Mc-Millan,04]
General LinearCamera (GLC) [Pajdla,02]
Admissible MapA
A
Goal
Stereo-vision betweenANY pair of
linear cameras
![Page 7: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/7.jpg)
OUTLINE
1. Stereo-vision with pinhole cameras
2. The admissible map model
3. Stereo-vision with linear cameras
![Page 8: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/8.jpg)
∼
![Page 9: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/9.jpg)
Inverse projection
![Page 10: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/10.jpg)
x
Inverse projection
![Page 11: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/11.jpg)
image of theblue line
x
Inverse projection
![Page 12: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/12.jpg)
stereo correspondance=
intersection ofinverse projections
u 7→ `
x
Inverse projection
![Page 13: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/13.jpg)
pinhole
captor lattice captor basis
world basisW
K
d
Normalized coordinates
translation (3)
rotation (3)
6 parameters
}
translation (2)
rotation (1)
5 parameters
scalings (2)
{normalized
basis
![Page 14: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/14.jpg)
Fundamental & Essential matrix
1u1
u2
Fundamental matrix
uT F u′ = 03× 3 matrix
Essential matrix
uT E u′ = 0E depends only on W
W
K
d
8 correspondances neededto determine F
6 correspondances neededto determine E
![Page 15: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/15.jpg)
OUR APPROACH
![Page 16: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/16.jpg)
OUR APPROACH
Camera=
bag of lines+
retina
![Page 17: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/17.jpg)
OUTLINE
1. Stereo-vision with pinhole cameras
2. The admissible map model
3. Stereo-vision with linear cameras
![Page 18: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/18.jpg)
OUTLINE
1. Stereo-vision with pinhole cameras
2. The admissible map model
3. Stereo-vision with linear cameras
“bag of lines” level
adding the retina
![Page 19: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/19.jpg)
Projective space
P3(R) ∼ set of lines of R4
passing through the origin
Homogeneous coordinates
[x0 : x1 : x2 : x3] ∼ [λx0 : λx1 : λx2 : λx3]
Intuition in the affine space R3
P3(R) = points + directions
[1 : x : y : z] [0 : u : v : w]
parallelism
angles
~u `
P
` ∩ P = {[~u]}
![Page 20: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/20.jpg)
Linear congruences
bundle
hyperboliccongruence
paraboliccongruence
ellipticcongruence
X-slitcamera
pencilcamera
linearobliquecamera
pinholecamera (no name)
degeneratecongruence
3D field of view order 1: x!7→ `
⇒ 2-parameter set of lines
Projectiveclassification
+(for almost
all x)
![Page 21: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/21.jpg)
Linear congruences
bundle
hyperboliccongruence
paraboliccongruence
ellipticcongruence
X-slitcamera
pencilcamera
linearobliquecamera
pinholecamera (no name)
degeneratecongruence
Ambiguitylocus
two lines one line
empty∅
a point
a plane
3D field of view order 1: x!7→ `
⇒ 2-parameter set of lines+
(for almostall x)
![Page 22: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/22.jpg)
Admissible maps
Idea: a linear map A that globallypreserves each line of the bag
`
xAx
A2x
A =
0BB@∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗
1CCA = R4×4
![Page 23: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/23.jpg)
Admissible maps
Idea: a linear map A that globallypreserves each line of the bag
`
xAx
A2x
⇒ A2x = λxAx+ µxx(for almost all x)
⇒ A2 = λA+ µId (linear algebra)
Definition: A has a minimal polynomial πA of degree 2.
A =
0BB@∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗
1CCA = R4×4
![Page 24: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/24.jpg)
Admissible maps
Idea: a linear map A that globallypreserves each line of the bag
`
xAx
A2x
L = {x ∨Ax for x not an eigenvector}
⇒ A2x = λxAx+ µxx(for almost all x)
⇒ A2 = λA+ µId (linear algebra)
Definition: A has a minimal polynomial πA of degree 2.
⇒ has order 1.
A =
0BB@∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗∗ ∗ ∗ ∗
1CCA = R4×4
Ambiguity locusof L
Union of eigenspacesof A
⊆
![Page 25: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/25.jpg)
Admissible mapsπA Eigenspaces Reduced form of A L
(X − α) (X − β) plane + point
[αI3 00 β
]
(X − α) (X − β) 2 lines
[αI2 00 βI2
]
(X − α)2 plane
αI2 0
0α 0λ α
(X − α)2 1 line
α 0λ α
0
0α 0µ α
∆ < 0 ∅
α −ββ α
0
0α −ββ α
Geometricmodel
(admissible maps)
(linear congruences)
⇔
Analyticalmodel
![Page 26: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/26.jpg)
class LinearCamera : public Camera {
〈 LinearCamera Constructor 〉float GenerateRay(Sample &s, Ray *r);
Transform AdMap;
Transform RasterToWorld;
Vector vview;
〈 Other Attributes 〉 };
Application: Ray-tracing (using pbrt)
![Page 27: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/27.jpg)
float GenerateRay(Sample &s,Ray *r) {
Point pras = Point(sample.imageX,sample.imageY,0);
Point pret = RasterToWorld(pras);
ray->o = pret;
ray->d = Normalize(AdMap(pret)-pret);
if (Dot(ray->d,vview) < 0) ray->d = -(ray->d);
〈 Setting Ray Time and Endpoints 〉 };
Application: Ray-tracing (using pbrt)
![Page 28: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/28.jpg)
OUTLINE
1. Stereo-vision with pinhole cameras
2. The admissible map model
3. Stereo-vision with linear cameras2 key ingredients:
? inverse projection
? normalized coordinates
![Page 29: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/29.jpg)
Inverse projection
∀x non-ambiguous x 7−→ x ∨ Ax ∈ Lwhere A is an admissible map for L
`
xx ∨Ax = [ξ0 : ξ1 : ξ2 : ξ3 : ξ4 : ξ5]
`
x y0
y1
y2
x =∑
i ui yi Inverse projection
πi : (ui) 7−→ (ξi)
![Page 30: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/30.jpg)
Inverse projection
πi(u) =
2X
i=0
uiyi
!∨
2X
i=0
uiAyi
!
πi(u) =
2Xi=0
ui2
2ζii +
X0≤i<j≤2
uiuj ζij
ζij = yi ∨Ayj + yj ∨Ayiwhere
eP = [ζ00, ζ01, ζ02, ζ11, ζ12, ζ22]
µ(u) = (u02, u0u1, u0u2, u1
2, u1u2, u22)T
(6× 6 matrix)
(6-vector)
πi(u) = P̃ µ(u)
![Page 31: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/31.jpg)
Fundamental matrix
ξ = [ξ0 : ξ1 : ξ2 : ξ3 : ξ4 : ξ5]T
ξ∗ = [ξ3 : ξ4 : ξ5 : ξ0 : ξ1 : ξ2]Tξ � ξ′ = ξT · (ξ′)∗
side-operator
` ∩ `′ 6= ∅ ⇔ ξ � ξ′ = 0
u u′
πi(u) πi(u′) Image coordinates u and u′ are
in stereo correspondance⇐⇒
πi(u)� πi(u′) = 0
![Page 32: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/32.jpg)
Fundamental matrix
u u′
πi(u) πi(u′) Image coordinates u and u′ are
in stereo correspondance⇐⇒
πi(u)� πi(u′) = 0
µ(u)T P̃T (P̃ ′)∗ µ(u′) = 0
6× 6 fundamental matrix
35 correspondances needed
![Page 33: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/33.jpg)
Normalized coordinates
Idea: the bag of lines is spanned by at most 4 lines4express πi in that “base” of lines
![Page 34: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/34.jpg)
Normalized coordinates
Idea: the bag of lines is spanned by at most 4 lines4express πi in that “base” of lines
Rp1
[p0+p3]p3
ζ1ζ0
ζ3
ζ2
p2
p0
X-slit camera
(p0, p1, p2, p3) = basis of the camera
(p0, p1, p2) = basis of the retina8>><>>:ζ0 = p0 ∨ p3
ζ1 = p1 ∨ (p0 + p3)ζ2 = p2 ∨ p3
ζ3 = p1 ∨ p2
basis ofthe bag of
lines
![Page 35: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/35.jpg)
Normalized coordinates
Idea: the bag of lines is spanned by at most 4 lines4express πi in that “base” of lines
Rp1
[p0+p3]p3
ζ1ζ0
ζ3
ζ2
p2
p0
X-slit camera
(p0, p1, p2, p3) = basis of the camera
(p0, p1, p2) = basis of the retina8>><>>:ζ0 = p0 ∨ p3
ζ1 = p1 ∨ (p0 + p3)ζ2 = p2 ∨ p3
ζ3 = p1 ∨ p2
basis ofthe bag of
lines
As = 12
(1 0 0 00 −1 0 00 0 1 02 0 0 −1
)P̃s × µs(u) =
0 −1 0 00 0 0 01 0 0 00 0 1 00 −1 0 00 0 0 1
×u0
2
u0u1
u0u2
u1u2
![Page 36: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/36.jpg)
Normalized coordinates
Rp1
ζ3 ζ ′2
ζ ′1
ζ0
[p2+ip1]
[p0+ip3]
p2
p0p3
R
ζ3
ζ0
p1
p2
p0
p3
[p0+p2]
[p1+p3]
ζ ′2
ζ1
Pencil camera
Linear oblique camera
P̃s × µs(u) =
0 0 1 00 0 0 01 0 0 00 0 1 00 −1 0 00 0 0 −1
×u0
2
u0u1
u0u2
u22
P̃s × µs(u) =
0 0 1 00 −1 0 01 0 0 00 0 1 00 −1 0 00 0 0 −1
×
u02
u0u1
u0u2
u12 + u2
2
P̃s and µs depend onlyon the type of the camera.
![Page 37: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/37.jpg)
Normalized coordinates
Also for pinhole cameras!
R p1
p3
ζ1
ζ0ζ2
p2 p0P̃s × µs(u) =
0 0 00 0 01 0 00 0 10 1 00 0 0
×u0
u1
u2
![Page 38: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/38.jpg)
Normalized coordinates
Calibration? position of the camera in the world → W
? position of the captor lattice in the retina → K
camera coordinatesin (p0, p1, p2, p3)
world coordinatesin (e0, e1, e2, e3)
invertible4× 4 matrix
retina coordinatesin (p0, p1, p2)
physical / pixelcoordinates
invertible3× 3 matrix
![Page 39: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/39.jpg)
Normalized coordinates
Calibration? position of the camera in the world → W
? position of the captor lattice in the retina → K
camera coordinatesin (p0, p1, p2, p3)
world coordinatesin (e0, e1, e2, e3)
invertible4× 4 matrix
retina coordinatesin (p0, p1, p2)
physical / pixelcoordinates
invertible3× 3 matrix
πi(u) = (Λ2W ) P̃s × µs(K−1u)
Λ2W is a 6× 6 matrix encoding the action of W on lines.
![Page 40: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/40.jpg)
Essential matrix
u u′
πi(u) πi(u′) Image coordinates u and u′ are
in stereo correspondance⇐⇒
πi(u)� πi(u′) = 0
µs(K−1u)T P̃T
s Λ2W T (Λ2W ′)∗ (P̃ ′s)∗ µs(K′−1u′) = 0
4× 4 essential matrix15 correspondances needed
![Page 41: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/41.jpg)
Euclidean framework
Exactly the same machinery
Need orthonormal normalized basis (p0, p1, p2, p3)
Only 6 correspondances needed to build E
Induce more involved intrinsic parameters
![Page 42: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/42.jpg)
6× 6 fundamental matrix (uncalibrated case)
4× 4 essential matrix (calibrated case)
between ANY pair of linear cameras
bag of lines+
retinaCamera =
{A
AG
L
LinearCongruences
GrassmannianSections
AdmissibleMaps
![Page 43: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/43.jpg)
FURTHER QUESTIONS
Do algebraic line congruencessuggest interesting imagingdevices? Can they admitanalogues of admissible maps?
What are the positionsof the retina that minimize
the distorsions of the image?
Can we get clear pictures fromour pencil camera?
[Pajdla et al.]
![Page 44: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/44.jpg)
Thank you!
[1] J. Ponce, What is a Camera?, CVPR’09
[2] G. B., X. Goaoc, J. Ponce, Admissible Linear Map Modelfor Linear Cameras, CVPR’10
[3] The “linear camera” model (in preparation)
![Page 45: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/45.jpg)
Pencil camera
our pencil camera
![Page 46: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/46.jpg)
Synthesis
projective euclidean euclideanlinear linear pinhole
camera camera camera
extrinsic parameters 15 6 6
intrinsic parameters 6 12 5
fundamental matrix 27 7
![Page 47: Bag of lines models - Center For Machine Perception (Cmp)cmp.felk.cvut.cz/cmp/events/colloquium-2010.10.21/batog...2010/10/21 · Guillaume BATOG INRIA Nancy Grand-Est project-team](https://reader033.fdocuments.us/reader033/viewer/2022052104/603fd13bba782d5feb5700f9/html5/thumbnails/47.jpg)
Geometric axiomatisation
Linearity on lines
[Veblen&Young,1910]
Initialisation
line pencil
degenerateregulus
for linear dependance
regulustwo lines
Heredity` depends linearly on L = {`1, ..., `k} iff∃ L→ `k+1 → · · · → `k+n = `
where → is a line pencil/regulus construction.
Exemples
bundle field(Need 2 line pencil steps)