• Aucun résultat trouvé

2 Fundamental matrix

N/A
N/A
Protected

Academic year: 2022

Partager "2 Fundamental matrix"

Copied!
3
0
0

Texte intégral

(1)

Vision 3D artificielle - Final exam (duration: 2h30)

P. Monasse, M. Aubry, and R. Marlet November 10th, 2015

You can choose to answer in French or English, at your convenience.

1 SVD

The two questions are independent.

1. LetE be a 3×3 real matrix. Show that:

(det(E) = 0

2EETE= tr(EET)E ⇔E’s singular values are (σ, σ,0) with σ≥0.

2. In the following, we use the Frobenius normkAk2 = tr(ATA). LetA be ann×nreal matrix of coefficientsaij.

(a) Show that kAk2=P

ija2ij.

(b) Show that this defines a matrix norm, indeed.

(c) Show that ifU is orthogonal,kU Ak=kAk.

(d) Show that

arg min

A0,det(A0)=0

kA0−Ak

can be computed simply through the SVD of A. Explain how and justify it.

2 Fundamental matrix

1. When trying to compute the fundamental matrix F with automatically matched feature points and RANSAC algorithm on the images of Fig- ure 1, we find a solution, but clicking points on the box do not show lines corresponding to matching points. Explain why.

1

(2)

Q

P

Figure 1: Top: original image pair and detected feature points. Bottom: two clicked points, one on the central left part of the ground (P) and one on the box (Q) with their computed epipolar lines. The one associated toQis not correct.

2. LetH be the homography relating a planar part of the scene in a stereo pair. Show that we can write the fundamental matrixF = [e]×H−1 with esome 3-vector.

3. Show that the data of two matched points outside the planar part of the scene are sufficient to determinee.

4. Propose an algorithm (in pseudo-code), variant of RANSAC, that may be able to handle correctly the problem of Figure 1.

3 Multi-view constraints

Suppose that a 3D point is seen in three views through pinhole cameras, at positionsx0= x10 x20 x30T

, x1,x2. We can write the trifocal constraints

[x1]× X

i

xi0Ti

!

[x2]×= 0

whereTi (i= 1,2,3) are 3×3 matrices. Show that among these 9 constraints on the coefficients ofTi, at most 4 are linearly independent equations.

2

(3)

4 Disparities estimation with graph-cut techniques

Given two rectified images, consider the problem of assigning a disparity to each pixel on each image line, in the set{dmin, . . . , dmax}. We consider the following strategy:

Initially consider all pixels as unassigned;

ford←dmin tod←dmax do

Build a graph whose cut best separates pixels into {D≤d, D>d}where D≤d are pixels with disparityd0≤dandD>d are pixels with

disparityd0 > d;

end

forall thepixelpdo

Assign topthe disparity max

d∈{dmin,...,dmax} p∈D≤d\D>d

d;

end

Questions: What are the advantages and disadvantages of such a strategy compared to the other techniques presented in the course, in terms of:

1. problem size,

2. well-definedness of the algorithm, 3. quality of the solution being modeled, 4. optimality of the solution reached.

3

Références

Documents relatifs

The fundamental matrix is estimated on the points belonging to the background; the correspondences on the tree on the foreground lead consequently to more uncertain epipolar lines.

The problem at hand is to estimate the fundamental matrix between two views, based on two lists of points of interest (POI, given with a local descriptor), one list for each of the

This is not true for ssGBLUP because new genomic relationships may modify old pedigree relationships; however, the ef- fect of genotypes of young animals on GEBV of other

Study the variations of the curve and plot it together with the tangent lines at some points.. Calculate the length of

Please attach a cover sheet with a declaration http://tcd-ie.libguides.com/plagiarism/declaration confirming that you know and understand College rules on plagiarism1. On the same

Let P be a convex/undamental polyhedron/or the discrete group G, where every point o/A(G) either is a point o/approximation or is a eueped parabolic fixed

S with geo- metrically connected fibres and smooth, geometrically connected base over a finite field k, and a section p of f , we can define an affine group scheme p rig 1 (X=S; p)

k-scheme X. the subgroup of the Galois group generated by the inertia groups over all components of D is abelian)... Then fi 1 D is an effective divisor on