11
Multi-view Geometry Readings: FP10 March 11, 2006 Multi-view Geometry and 3D We have 2 eyes , thus we see 3D Using multiple views allows inference of the 3 rd dimension How to see in 3D (Using geometry…) Find corresponding features Triangulate and reconstruct depth Correspondence Last Lecture: Now: Outline Multi-view Geometry

11 Multi-view Geometry - web.media.mit.edu

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: 11 Multi-view Geometry - web.media.mit.edu

1

Multi-view Geometry

Readings: FP10

March 11, 2006

Multi-view Geometry and 3D• We have 2 eyes , thus we see 3D

• Using multiple views allows inference of the 3rd

dimension

How to see in 3D(Using geometry…)

• Find corresponding features• Triangulate and reconstruct depth

Correspondence

Last Lecture:

Now:

Outline Multi-view Geometry

Page 2: 11 Multi-view Geometry - web.media.mit.edu

2

Multi-view Geometry Multi-view Geometry

Multi-view Geometry Multi-view Geometry

Multi-view Geometry Stereo Constraints

Page 3: 11 Multi-view Geometry - web.media.mit.edu

3

Stereo Constraints Epipolar Line

Epipolar Constraint Epipolar Constraint

P

p p’

O O’

From Geometry to Algebra

P

p p’

O O’

P

O’O

p p’

p p’ ( )[ ] 0=+×⋅ p'ttp R

( )[ ] 0=×⋅ p'tp R

[ ] 0=×⋅ '' pOOO'Op

Page 4: 11 Multi-view Geometry - web.media.mit.edu

4

P

O’O

p p’

p p’ ( )[ ] 0=+×⋅ p'ttp R

( )[ ] 0=×⋅ p'tp R

[ ] 0=×⋅ '' pOOO'Op

Matrix Form

Review: Matrix Form of the Cross Product Review: Matrix Form of the Cross Product

Review: Matrix Form of the Cross Product Matrix Form

Page 5: 11 Multi-view Geometry - web.media.mit.edu

5

Matrix Form Essential Matrix

Epipolar Line Constraint Essential Matrix – Instantaneous case• For small motion given translation and rotation

velocity:

ε

Focus of Expansion for Translating Camera

FOE for Translating Camera

Page 6: 11 Multi-view Geometry - web.media.mit.edu

6

What if calibration is unknown? Fundamental Matrix

Estimating the Fundamental Matrix Estimating the Fundamental Matrix

8 Point Algorithm 8 Point Algorithm

Page 7: 11 Multi-view Geometry - web.media.mit.edu

7

Improved 8 Point Algorithm Normalizing the Input Data

Stereo Constraints

Given p1, p2, in the left and right image where is p3?

p1 p2p3

?

Page 8: 11 Multi-view Geometry - web.media.mit.edu

8

Three Essential Matrices

p1 p2p3

?

Trinocular Epipolar Geometry

Trinocular Epipolar Geometry

p1 is at the intersection of the epipolar lines associated with p2 and p3

Given p2 and p2, p1 is determined!

Recall: Epipolar Line Constraint

0=lpT

Three Essential Matrices

02121 =pp εT01313 =pp εT

From lecture 3:

M

Page 9: 11 Multi-view Geometry - web.media.mit.edu

9

C

From lecture 3:

1zyx

W

W

W

From lecture 3:

Trifocal Line Constraint Trifocal Line Constraint

Trifocal Line Constraint

Page 10: 11 Multi-view Geometry - web.media.mit.edu

10

The Trifocal Tensor Trifocal Line Constraint

Line Transfer Uncalibrated Case

Quadrifocal Geometry

Page 11: 11 Multi-view Geometry - web.media.mit.edu

11

Trifocal Constraint with Noise Project• Final project may be:

– An original implementation of a new or published idea

• A project proposal not longer than two pages must be submitted by April 1.

ProjectMarch 24: - Project Previews / Brainstorming3-5 minute presentation describing

– Specific Project Idea– Your research, or thesis proposal

(if it relates to vision)– Paper you are interested in and may form the basis

of a project