CS 294-167 Spring 2022

Geometry and Learning for 3D Vision

Yi Ma

UC Berkeley

Invitation to 3D vision
 Lecture 4

Two-View Geometry

Invitation to 3D vision
 General Formulation

Given two views of the scene,

recover the unknown camera
pose and 3D scene structure

MASKS © 2004 Invitation to 3D vision

 Calibrated Pinhole Camera Model

• 3D points

• Image points

• Perspective Projection

• Rigid Body Motion

• Rigid Body Motion + Projective projection

MASKS © 2004 Invitation to 3D vision

 Rigid Body Motion – Two views

MASKS © 2004 Invitation to 3D vision

 3D Structure and Motion Recovery

Euclidean transformation

measurements unknowns

Find such Rotation and Translation and Depth that

the reprojection error is minimized
Two views ~ 200 points
6 unknowns – Motion 3 Rotation, 3 Translation
- Structure 200x3 coordinates
- a universal scale
Difficult nonlinear optimization problem

MASKS © 2004 Invitation to 3D vision

 Epipolar Geometry

Image
correspondences

b b
<latexit sha1_base64="Gh1I22nGyRS8Kf19jAqnQXFZe4M=">AAACMHicbVDLSsNAFJ34rPUVdelmsAiuSlJE3QhFF7qs0hc0IUwmk3bo5MHMRC0hn+TGT9GNgiJu/QonbYTaemDgcM65zL3HjRkV0jDetIXFpeWV1dJaeX1jc2tb39ltiyjhmLRwxCLedZEgjIakJalkpBtzggKXkY47vMz9zh3hgkZhU45iYgeoH1KfYiSV5OhXFlNhDzk16556ZIBk2sysAMmB66cPmVOD5/A3YsKpzC2cSpnQ0StG1RgDzhOzIBVQoOHoz5YX4SQgocQMCdEzjVjaKeKSYkayspUIEiM8RH3SUzREARF2Oj44g4dK8aAfcfVCCcfq9ESKAiFGgauS+ZJi1svF/7xeIv0zO6VhnEgS4slHfsKgjGDeHvQoJ1iykSIIc6p2hXiAOMJSdVxWJZizJ8+Tdq1qnlSNm+NK/aKoowT2wQE4AiY4BXVwDRqgBTB4BC/gHXxoT9qr9ql9TaILWjGzB/5A+/4BuL2p+Q==</latexit>

2 T x2 = 1 T Rx1

b T b
<latexit sha1_base64="Gn8yVz2qjouJq4m6u1/8Fi86Psk=">AAACU3icbVFNS8MwGE7r1DmdVj16CQ7B02iHqBdh6MXjlH3BWkuapls0/SBJ1VH6H0Xw4B/x4kHTrYe5+ULg4fkgb554CaNCmuanpq9V1jc2q1u17Z367p6xf9AXccox6eGYxXzoIUEYjUhPUsnIMOEEhR4jA+/pptAHz4QLGkddOU2IE6JxRAOKkVSUazya8AraIZITL8hec7f10IU2U3kfuS37hfpkgmTWzRctRaK0WMtZuJC5XxQt6BoNs2nOBq4CqwQNUE7HNd5tP8ZpSCKJGRJiZJmJdDLEJcWM5DU7FSRB+AmNyUjBCIVEONmskxyeKMaHQczViSScsYuJDIVCTENPOYslxbJWkP9po1QGl05GoySVJMLzi4KUQRnDomDoU06wZFMFEOZU7QrxBHGEpfqGmirBWn7yKui3mtZ507w7a7Svyzqq4Agcg1NggQvQBregA3oAgzfwBX40oH1o37quV+ZWXSszh+DP6PVf5Aazag==</latexit>

0 = xT2 2 T x2 = 1 x2 T Rx1

• Algebraic Elimination of Depth [Longuet-Higgins ’81]:

• Essential matrix
MASKS © 2004 Invitation to 3D vision
 Epipolar Geometry

• Epipolar lines

• Epipoles

Image
correspondences

MASKS © 2004 Invitation to 3D vision

 Characterization of the Essential Matrix

n o
. b
<latexit sha1_base64="wxOjPEs+3U05v770ntd1F/ARAz4=">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</latexit>

E = T R | R 2 SO(3), T 2 <3 ⇢ R3⇥3

• Essential matrix Special 3x3 matrix

Theorem 1 (Essential Matrix Characterization)

A non-zero matrix is an essential matrix iff its SVD:
satisfies: with and
and

MASKS © 2004 Invitation to 3D vision

 Estimating the Essential Matrix

• Estimate Essential matrix :

• Decompose Essential matrix into

• Given n pairs of image correspondences:

Find such Rotation and Translation that the epipolar error is minimized
<latexit sha1_base64="MBZ8/a9AYA732TfPOuWJaA5J+Vc=">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</latexit>

n
X jT j 2
min x2 Ex1
E
j=1
• Space of all Essential Matrices is 5 dimensional
• 3 Degrees of Freedom – Rotation
• 2 Degrees of Freedom – Translation (up to scale !)

MASKS © 2004 Invitation to 3D vision

 Pose Recovery from the Essential Matrix

Essential matrix
Theorem 2 (Pose Recovery)
There are two relative poses with and
corresponding to a non-zero matrix essential matrix.

Twisted pair ambiguity:

MASKS © 2004 Invitation to 3D vision

 Estimating Essential Matrix

• Denote

• Rewrite

• Collect constraints from all points

MASKS © 2004 Invitation to 3D vision

 Estimating Essential Matrix

Solution
• Eigenvector associated with the smallest eigenvalue of
• if , degenerate configuration.

Projection onto the Essential Space

Theorem 3 (Projection to Essential Manifold)

If the SVD of a matrix is given by
then the essential matrix which minimizes the Frobenius
distance is given by with

MASKS © 2004 Invitation to 3D vision

 Two View 8-point Linear Algorithm

• Solve the LLSE problem:

<latexit sha1_base64="MBZ8/a9AYA732TfPOuWJaA5J+Vc=">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</latexit>

n
X 2
min xjT
2 Ex j
1
E
j=1
followed by projection
• Project onto the essential manifold:
SVD:

E is 5 diml. sub. mnfld. in

• Recover the unknown pose:

MASKS © 2004 Invitation to 3D vision

 Pose Recovery from Essential Martrix

• There are exactly two pairs corresponding to each

essential matrix .
• There are also two pairs corresponding to each
essential matrix .

• Positive depth constraint - used to disambiguate the

physically impossible solutions

• Translation has to be non-zero

• Points have to be in general position

- degenerate configurations – planar points
- quadratic surface
• Linear 8-point algorithm
• Nonlinear 5-point algorithms yield up to 10 solutions (Kruppa)

MASKS © 2004 Invitation to 3D vision

 3D Structure Recovery

• Eliminate one of the scales

• Solve LLSE problem

If the configuration is non-critical, the Euclidean structure of then points

and motion of the camera can be reconstructed up to a universal scale.

MASKS © 2004 Invitation to 3D vision

 3D Incidence (Rank) Constraint

bRxj = 0.
<latexit sha1_base64="CzRUF5PsmBB1xom1PhxDOulFPj4=">AAACGXicbVDLSsNAFJ3UV62vqks3g0VwFZIi6kYounFZpS9oYphMJu20kwczE7WE/IYbf8WNC0Vc6sq/cdoG1NYDFw7n3Mu997gxo0IaxpdWWFhcWl4prpbW1jc2t8rbOy0RJRyTJo5YxDsuEoTRkDQllYx0Yk5Q4DLSdocXY799S7igUdiQo5jYAeqF1KcYSSU5ZcMKkOy7Pry/SQeNzKlC6456pI9k2siu4Y87cEx4Bg3dKVcM3ZgAzhMzJxWQo+6UPywvwklAQokZEqJrGrG0U8QlxYxkJSsRJEZ4iHqkq2iIAiLsdPJZBg+U4kE/4qpCCSfq74kUBUKMAld1ji8Vs95Y/M/rJtI/tVMaxokkIZ4u8hMGZQTHMUGPcoIlGymCMKfqVoj7iCMsVZglFYI5+/I8aVV181g3ro4qtfM8jiLYA/vgEJjgBNTAJaiDJsDgATyBF/CqPWrP2pv2Pm0taPnMLvgD7fMbi+qfVA==</latexit>

xjT
2 T 1

MASKS © 2004 Invitation to 3D vision

 Example- Two-view Correspondence

Point Feature Matching

MASKS © 2004 Invitation to 3D vision

 Example – Use of Epipolar Geometry

Camera Pose
and
Sparse Structure Recovery

MASKS © 2004 Invitation to 3D vision

 Two-view Geometry – Planar Case

• Plane in first camera coordinate frame

N
<latexit sha1_base64="zRC70pPimSkX3nTZIRgTiyBN/wg=">AAAB6HicbVBNS8NAEJ34WetX1aOXxSJ4KomIeix68SQt2A9oQ9lsJ+3azSbsboQS+gu8eFDEqz/Jm//GbZuDtj4YeLw3w8y8IBFcG9f9dlZW19Y3Ngtbxe2d3b390sFhU8epYthgsYhVO6AaBZfYMNwIbCcKaRQIbAWj26nfekKleSwfzDhBP6IDyUPOqLFS/b5XKrsVdwayTLyclCFHrVf66vZjlkYoDRNU647nJsbPqDKcCZwUu6nGhLIRHWDHUkkj1H42O3RCTq3SJ2GsbElDZurviYxGWo+jwHZG1Az1ojcV//M6qQmv/YzLJDUo2XxRmApiYjL9mvS5QmbE2BLKFLe3EjakijJjsynaELzFl5dJ87ziXVbc+kW5epPHUYBjOIEz8OAKqnAHNWgAA4RneIU359F5cd6dj3nripPPHMEfOJ8/qDeM1w==</latexit>

Image
correspondences

Planar homography

Linear mapping relating two

corresponding planar points
in two views

MASKS © 2004 Invitation to 3D vision

 Two-view Geometry – Role of Homography

T
<latexit sha1_base64="bKB8q9871HRtxcnrsozaeRBvsP4=">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</latexit>

c2 Hx1 ,
`2 ⇠ x `1 ⇠ H `2

MASKS © 2004 Invitation to 3D vision

 Decomposition of H

• Algebraic elimination of depth

• can be estimated linearly
• Normalization of
• Decomposition of H into 4 solutions

MASKS © 2004 Invitation to 3D vision

 Motion and Pose Recovery from Homography

• Given at least 4 point correspondences

s
Compute an approximation of the homography matrix HL
<latexit sha1_base64="CSH9Y/rKPEf4Kj7jH7M/wXqEq6o=">AAAB7HicbVA9SwNBEJ3zM8avqKXNYhCswp2IWgZtUlhE8JJAcoa9zVyyZG/v2N0TQshvsLFQxNYfZOe/cZNcoYkPBh7vzTAzL0wF18Z1v52V1bX1jc3CVnF7Z3dvv3Rw2NBJphj6LBGJaoVUo+ASfcONwFaqkMahwGY4vJ36zSdUmifywYxSDGLalzzijBor+bXu3aPulspuxZ2BLBMvJ2XIUe+Wvjq9hGUxSsME1brtuakJxlQZzgROip1MY0rZkPaxbamkMepgPDt2Qk6t0iNRomxJQ2bq74kxjbUexaHtjKkZ6EVvKv7ntTMTXQdjLtPMoGTzRVEmiEnI9HPS4wqZESNLKFPc3krYgCrKjM2naEPwFl9eJo3zindZce8vytWbPI4CHMMJnIEHV1CFGtTBBwYcnuEV3hzpvDjvzse8dcXJZ47gD5zPH3YLjnU=</latexit>

as nullspace of
HLs = 0 the rows of are
<latexit sha1_base64="mdoEFBQFxY1gPtgxWrpc1SZscYU=">AAAB9XicbVA9SwNBEJ2LXzF+RS1tFoNgFe5E1EYI2qSwiGA+ILmEvc1esmR379jdU8KR/2FjoYit/8XOf+MmuUITHww83pthZl4Qc6aN6347uZXVtfWN/GZha3tnd6+4f9DQUaIIrZOIR6oVYE05k7RumOG0FSuKRcBpMxjdTv3mI1WaRfLBjGPqCzyQLGQEGyt1O2TIULV319XoGrm9YsktuzOgZeJlpAQZar3iV6cfkURQaQjHWrc9NzZ+ipVhhNNJoZNoGmMywgPatlRiQbWfzq6eoBOr9FEYKVvSoJn6eyLFQuuxCGynwGaoF72p+J/XTkx45adMxomhkswXhQlHJkLTCFCfKUoMH1uCiWL2VkSGWGFibFAFG4K3+PIyaZyVvYuye39eqtxkceThCI7hFDy4hApUoQZ1IKDgGV7hzXlyXpx352PemnOymUP4A+fzB3LqkSw=</latexit>

• Normalize the homography matrix

• Decompose the homography matrix

• Select two physically possible solutions imposing

positive depth constraint

MASKS © 2004 Invitation to 3D vision

 Homography to Essential Matrix (Two Point)

T ⇠ `b12 `22
<latexit sha1_base64="c2Hk7BRgznF4OPCWfwjyjEZpsWE=">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</latexit>

E = TbH,

Two more
generic points

MASKS © 2004 Invitation to 3D vision

 Essential Matrix to Homography (Three Points)

H = TbT E + T v T ,
<latexit sha1_base64="hW89eBqp9dt5Au70p6PvXH1Tadg=">AAACHnicbVBNS8NAEN3U7/pV9ehlsAiCUhK/L0JRhB5VUi00bdlstu3iZhN3N5US+ku8+Fe8eFBE8KT/xm3tQVsfDDzem2Fmnh9zprRtf1mZicmp6ZnZuez8wuLScm5l9VpFiSS0TCIeyYqPFeVM0LJmmtNKLCkOfU5v/Nuzvn/ToVKxSLi6G9NaiFuCNRnB2kiN3EEJTsC7ZwFtY526vboL57ANLnTq7g54dwkOoAMeE+CFWLd9P73q1fcaubxdsAeAceIMSR4NcdHIfXhBRJKQCk04Vqrq2LGupVhqRjjtZb1E0RiTW9yiVUMFDqmqpYP3erBplACakTQlNAzU3xMpDpXqhr7p7N+oRr2++J9XTXTzuJYyESeaCvKzqJlw0BH0s4KASUo07xqCiWTmViBtLDHRJtGsCcEZfXmcXO8WnMOCfbmfL54O45hF62gDbSEHHaEiKqELVEYEPaAn9IJerUfr2Xqz3n9aM9ZwZg39gfX5DWAtoDQ=</latexit>

v 2 R3

Three points on the plane

<latexit sha1_base64="jFF7TEL3H1m1GCqf3ZvUTGdaGeg=">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</latexit>

c
xj2 TbT E + T v T xj1 = 0, j = 1, 2, 3.

MASKS © 2004 Invitation to 3D vision

 Example

H
<latexit sha1_base64="NgWE8hSOBo5S/Gqp8RpRWjMrVCg=">AAAB6HicbVBNS8NAEJ34WetX1aOXxSJ4KomIeix66bEF+wFtKJvtpF272YTdjVBCf4EXD4p49Sd589+4bXPQ1gcDj/dmmJkXJIJr47rfztr6xubWdmGnuLu3f3BYOjpu6ThVDJssFrHqBFSj4BKbhhuBnUQhjQKB7WB8P/PbT6g0j+WDmSToR3QoecgZNVZq1Pqlsltx5yCrxMtJGXLU+6Wv3iBmaYTSMEG17npuYvyMKsOZwGmxl2pMKBvTIXYtlTRC7WfzQ6fk3CoDEsbKljRkrv6eyGik9SQKbGdEzUgvezPxP6+bmvDWz7hMUoOSLRaFqSAmJrOvyYArZEZMLKFMcXsrYSOqKDM2m6INwVt+eZW0LivedcVtXJWrd3kcBTiFM7gAD26gCjWoQxMYIDzDK7w5j86L8+58LFrXnHzmBP7A+fwBnx+M0Q==</latexit>

MASKS © 2004 Invitation to 3D vision

 Example: Image Stitching

H
<latexit sha1_base64="NgWE8hSOBo5S/Gqp8RpRWjMrVCg=">AAAB6HicbVBNS8NAEJ34WetX1aOXxSJ4KomIeix66bEF+wFtKJvtpF272YTdjVBCf4EXD4p49Sd589+4bXPQ1gcDj/dmmJkXJIJr47rfztr6xubWdmGnuLu3f3BYOjpu6ThVDJssFrHqBFSj4BKbhhuBnUQhjQKB7WB8P/PbT6g0j+WDmSToR3QoecgZNVZq1Pqlsltx5yCrxMtJGXLU+6Wv3iBmaYTSMEG17npuYvyMKsOZwGmxl2pMKBvTIXYtlTRC7WfzQ6fk3CoDEsbKljRkrv6eyGik9SQKbGdEzUgvezPxP6+bmvDWz7hMUoOSLRaFqSAmJrOvyYArZEZMLKFMcXsrYSOqKDM2m6INwVt+eZW0LivedcVtXJWrd3kcBTiFM7gAD26gCjWoQxMYIDzDK7w5j86L8+58LFrXnHzmBP7A+fwBnx+M0Q==</latexit>

MASKS © 2004 Invitation to 3D vision

 Example: Rectifying Planar Regular Textures
green window

rectified green window

MASKS © 2004 Invitation to 3D vision

 Special Homography: Rotation Case

• Two view related by rotation only

• Mapping to a reference view

• Mapping to a cylindrical surface

MASKS © 2004 Invitation to 3D vision

 Special Homography: Reflection Case

Symmetry on object
2 (1)
1(2)

4 (3)
3(4)

Virtual camera-camera

MASKS © 2004 Invitation to 3D vision

 Special Homography: Translation Case

MASKS © 2004 Invitation to 3D vision

 Application: Image Editing

MASKS © 2004 Invitation to 3D vision

 Application: Image Editing

MASKS © 2004 Invitation to 3D vision

 Application: Augmented Reality

MASKS © 2004 Invitation to 3D vision

 Motion and Structure Recovery – Two Views

• Two discrete views – general motion, general structure

n Estimate the essential matrix (from 8 points)

n Decompose the essential matrix E

n Impose positive depth constraint

n Recover 3D structure

• Two discrete views – general motion, planar structure

n Estimate the planar homography (from 4 points)

n Normalize and decompose H

n Recover 3D structure and camera pose

n Special homography from a single view (of symmetry)

• Corresponding continuous-motion cases (Section 5.4)

MASKS © 2004 Invitation to 3D vision

Learning for 3D Geometry: How to detect planes?