Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
29 views

14.camera Calibration

This document discusses camera calibration and omnidirectional camera calibration. It first describes common camera models and what camera calibration is, which is estimating camera parameters from known 3D object points and their corresponding 2D image points. It then summarizes Tsai's and Zhang's camera calibration methods which use a homography between the image plane and world plane to retrieve intrinsic and extrinsic parameters. Finally, it outlines a flexible technique for omnidirectional camera calibration that models the camera using a Taylor expansion and estimates calibration parameters including the expansion coefficients, rotation, and translation through a linear least squares problem.

Uploaded by

Usman Maqbool
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
29 views

14.camera Calibration

This document discusses camera calibration and omnidirectional camera calibration. It first describes common camera models and what camera calibration is, which is estimating camera parameters from known 3D object points and their corresponding 2D image points. It then summarizes Tsai's and Zhang's camera calibration methods which use a homography between the image plane and world plane to retrieve intrinsic and extrinsic parameters. Finally, it outlines a flexible technique for omnidirectional camera calibration that models the camera using a Taylor expansion and estimates calibration parameters including the expansion coefficients, rotation, and translation through a linear least squares problem.

Uploaded by

Usman Maqbool
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 25

Camera calibration &

Omnidirectional camera calibration

Ph.D. student
Chang-Ryeol Lee

November 26, 2013


Contents
• Camera model
• What is camera calibration?
• Camera calibration
• Tsai’s and Zhang’s methods

• Omnidirectional camera
• Camera model
• Calibration: A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
Camera model

* Prof. Kuk-Jin Yoon’s slides on High-level Image Understanding and Processing


What is calibration?

* Prof. Kuk-Jin Yoon’s slides on High-level Image Understanding and Processing


Camera model

* Prof. Kuk-Jin Yoon’s slides on High-level Image Understanding and Processing


What is camera calibration?
What is camera calibration?
Camera calibration
• We know positions of pattern corners with respect to world and
image coordinate
3D corners
2D corners 0 0 0 
 433 445 457
xI  
 XW  0, 3, 6  (cm)

 ( pixel )
 245 230 215  0 0 0 

y
z

Grid size:3cm
{World} x
{Image}
Camera calibration
• Camera parameters
• Intrinsic parameter K
• Focal length, skew, principal point
• Extrinsic parameter P
• Rotation, translation

x kI  KPXWk
 fx s px   r11 r12 r13 t1 
x kI   0 fy p y   r21 r22 r23 t2  XWk
 0 0 1   r31 r32 r33 t3 
Tsai’s and Zhang’s methods
• Retrieve not only the intrinsic but also extrinsic parameters
• Set their Z-coordinate to 0
• because the 3D points on the object are coplanar.

X
X
Y
m ≅ 𝐊 𝐑T | −𝐑T t M = 𝐊 r1 r2 r3 t ′ = 𝐊 r1 r2 t ′ Y
0
1
1

• A homography between m and M

𝐇 = h1 h2 h3 = λ𝐊 r1 r2 t ′
Tsai’s and Zhang’s methods
• Solving for Homography
Tsai’s and Zhang’s methods
• Any rotation matrix is an orthonormal matrix

r1 r2 t ′ = 𝐊 −1 h1 h2 h3
r1 𝑇 r2 = (𝐊 −1 h1 )𝑇 𝐊 −1 h2 = h2 𝑇 𝐊 −𝑇 𝐊 −1 h2 = 0
r1 𝑇 r1 = r2 𝑇 r2 = (𝐊 −1 h1 )𝑇 𝐊 −1 h1 = h1 𝑇 𝐊 −𝑇 𝐊 −1 h1 = h2 𝑇 𝐊 −𝑇 𝐊 −1 h2

• Define the symmetric matrix B as


- 6 unknown variable
𝐵11 𝐵12 𝐵13
𝐁 = 𝐊 −𝑇 𝐊 −1 = 𝐵12 𝐵22 𝐵23 = 0
𝐵13 𝐵23 𝐵33
Tsai’s and Zhang’s methods
• One image offers two linear equations in elements of B

h1 𝑇 𝐁h1 = h2 𝑇 𝐁h2 , h2 𝑇 𝐁h2 = 0

• We can get intrinsic parameter through 3 or more images


• because unknown variable(𝐵11 ⋯ 𝐵33 ) is 6.
Tsai’s and Zhang’s methods
• Derive equations for the internal parameters of K

𝐵11 𝐵12 𝐵13 𝛼𝑥 𝑠 𝑝𝑥 −𝑇 𝛼𝑥 𝑠 𝑝𝑥 −1

𝐵12 𝐵22 𝐵23 = 0 𝛼𝑦 𝑝𝑦 0 𝛼𝑦 𝑝𝑦


𝐵13 𝐵23 𝐵33 0 0 0 0 0 0
𝐁 = 𝐊 −𝑇 𝐊 −1

𝐵12 𝐵13 − 𝐵13 𝐵23 𝐵13 2 + 𝑢𝑦 (𝐵12 𝐵13 − 𝐵11 𝐵23 )


𝑢𝑦 = λ = 𝐵33 −
𝐵11 𝐵22 − 𝐵12 2 𝐵11
λ λ𝐵11
𝛼𝑥 = 𝛼𝑦 =
𝐵11 𝐵11 𝐵22 − 𝐵12 2

𝐵12 𝛼𝑥 2 𝛼𝑦 𝑠𝑢𝑦 𝛼2
𝑠=− 𝑢𝑥 = − 𝐵13
λ 𝛼 λ
Tsai’s and Zhang’s methods
• After intrinsic parameters, the extrinsics readily follow

r1 r2 t ′ = 𝐊 −1 h1 h2 h3

r1 = μ𝐊 −1 h1 r2 = μ𝐊 −1 h2 t = μ𝐊 −1 h3

1
r3 = r1 × r2 μ =
𝐊 −1 h1
Omnidirectional camera

Dioptric (fish-eye lens)

Catadioptric
(camera with mirror)
Omnidirectional camera
• Camera model

Catadioptric case Dioptric case


A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• Don’t care whether catadioptric or dordioptric
• General method even NASA use this currently
• Key idea
• Camera modeling by Taylor expansion

 u 
 x    v    R t  X  PX
 f (  ) 

f (  )  a0  a1  a2  2   aN  N
A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• No focal length, skew, principal point
• Calibration parameters
• Taylor expansion coefficients a
• Rotation R
• Translation t

 u 
 x    v    R t  X  PX
 f (  ) 

f (  )  a0  a1  a2  2   aN  N
A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• Set their Z-coordinate to 0 by planarity of pattern
A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• Cross product of a corner point

make three equations


A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• The 3rd equation can be linearly expressed on unknowns

• We know X j , Yj , u j , v j

• Homogeneous linear least square problem is solved by


SVD
A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• 𝑟3 is uniquely determined by 𝑟1 , 𝑟2
• Extrinsic parameter is determined except for 𝑡3

• The 1st, 2nd equations are used to determine t3 , a  [a0 , a1 , a2 aN ]T

, where f (  )  a0  a1  a2  2   aN  N
A flexible Technique for Accurate Omnidirectional
Camera Calibration and structure from motion
• Equations stacked with K observations of pattern is
expressed linearly

1st

Kth

where

• Unknown parameter 𝑡3 , 𝐚 are estimated


Question or Comments ?

You might also like