Projective geometry ECE 847: Digital Image Processing Stan Birchfield Clemson University.

Slides:



Advertisements
Similar presentations
Epipolar Geometry.
Advertisements

Computer Vision, Robert Pless
Invariants (continued).
1 A camera is modeled as a map from a space pt (X,Y,Z) to a pixel (u,v) by ‘homogeneous coordinates’ have been used to ‘treat’ translations ‘multiplicatively’
3D reconstruction.
Computer Graphics Lecture 4 Geometry & Transformations.
Primitives Behaviour at infinity HZ 2.2 Projective DLT alg Invariants
Recovery of affine and metric properties from images in 2D Projective space Ko Dae-Won.
Geometry 2: A taste of projective geometry Introduction to Computer Vision Ronen Basri Weizmann Institute of Science.
Camera Calibration. Issues: what are intrinsic parameters of the camera? what is the camera matrix? (intrinsic+extrinsic) General strategy: view calibration.
Camera Models A camera is a mapping between the 3D world and a 2D image The principal camera of interest is central projection.
2-D Geometry. The Image Formation Pipeline Outline Vector, matrix basics 2-D point transformations Translation, scaling, rotation, shear Homogeneous.
Lecture 11: Transformations CS4670/5760: Computer Vision Kavita Bala.
Multiple View Geometry Projective Geometry & Transformations of 2D Vladimir Nedović Intelligent Systems Lab Amsterdam (ISLA) Informatics Institute,
1 Basic geometric concepts to understand Affine, Euclidean geometries (inhomogeneous coordinates) projective geometry (homogeneous coordinates) plane at.
The 2D Projective Plane Points and Lines.
Projective 2D & 3D geometry course 2
Projective geometry- 2D Acknowledgements Marc Pollefeys: for allowing the use of his excellent slides on this topic
Epipolar geometry. (i)Correspondence geometry: Given an image point x in the first view, how does this constrain the position of the corresponding point.
Uncalibrated Geometry & Stratification Sastry and Yang
Epipolar Geometry and the Fundamental Matrix F
Projective geometry Slides from Steve Seitz and Daniel DeMenthon
3D reconstruction of cameras and structure x i = PX i x’ i = P’X i.
Multiple View Geometry
Multiple View Geometry Marc Pollefeys University of North Carolina at Chapel Hill Modified by Philippos Mordohai.
3D photography Marc Pollefeys Fall 2007
Projective Geometry and Camera model Class 2
6/28/2015©Zachary Wartell Homogenous Coordinates and Projective Geometry (Crudely Speaking) Revision 1.2 Copyright Zachary Wartell, University of North.
The Pinhole Camera Model
Projected image of a cube. Classical Calibration.
Lec 21: Fundamental Matrix
Projective 2D geometry course 2 Multiple View Geometry Comp Marc Pollefeys.
Automatic Camera Calibration
Computer vision: models, learning and inference
1 Homogeneous Coordinates and Transformation. 2 Line in R 2 General line equation Normalize: Distance to origin (projection along n) n (x,y) For any two.
875: Recent Advances in Geometric Computer Vision & Recognition Jan-Michael Frahm Fall 2011.
Homogeneous Coordinates (Projective Space) Let be a point in Euclidean space Change to homogeneous coordinates: Defined up to scale: Can go back to non-homogeneous.
Autonomous Navigation for Flying Robots Lecture 2.2: 2D Geometry
Projective Geometry. Projection Vanishing lines m and n.
Imaging Geometry for the Pinhole Camera Outline: Motivation |The pinhole camera.
University of Palestine Faculty of Applied Engineering and Urban Planning Software Engineering Department Introduction to computer vision Chapter 2: Image.
Projective geometry ECE 847: Digital Image Processing Stan Birchfield Clemson University.
Objects at infinity used in calibration
Geometric Camera Models
Two-view geometry Epipolar geometry F-matrix comp. 3D reconstruction
Multi-linear Systems and Invariant Theory in the Context of Computer Vision and Graphics CS329 Amnon Shashua.
A Flexible New Technique for Camera Calibration Zhengyou Zhang Sung Huh CSPS 643 Individual Presentation 1 February 25,
1 Chapter 2: Geometric Camera Models Objective: Formulate the geometrical relationships between image and scene measurements Scene: a 3-D function, g(x,y,z)
Projective Geometry Hu Zhan Yi. Entities At Infinity The ordinary space in which we lie is Euclidean space. The parallel lines usually do not intersect.
Computer vision: models, learning and inference M Ahad Multiple Cameras
Lecture 15: Transforms and Alignment CS4670/5670: Computer Vision Kavita Bala.
Projective 2D geometry course 2 Multiple View Geometry Comp Marc Pollefeys.
Homogeneous Coordinates They work, but where do they come from? Jonathan Senning
CS682, Jana Kosecka Rigid Body Motion and Image Formation Jana Kosecka
Calibration ECE 847: Digital Image Processing Stan Birchfield Clemson University.
Projective geometry ECE 847: Digital Image Processing Stan Birchfield
Homogeneous Coordinates (Projective Space)
Two-view geometry Computer Vision Spring 2018, Lecture 10
Epipolar geometry.
CS Visual Recognition Projective Geometry Projective Geometry is a mathematical framework describing image formation by perspective camera. Under.
Homogeneous Coordinates and Transformation
Computer Graphics Recitation 12.
2-D Geometry.
Lecture 3: Camera Rotations and Homographies
Geometric Camera Models
2D transformations (a.k.a. warping)
Announcements list – let me know if you have NOT been getting mail
컴퓨터 그래픽스를 위한 기하학적 기초 (Line Geometry for Computer Graphics)
The Pinhole Camera Model
Presentation transcript:

Projective geometry ECE 847: Digital Image Processing Stan Birchfield Clemson University

Lines A line in 2D is described by two parameters: But vertical lines? Only two parameters are sufficient, but requires nonlinear formulation: slopey-intercept almost ^

Lines A better parameterization can represent all lines: Here the line is represented by 3 parameters: But nonzero scalar multiple does not change the equation: So we have only 2 degrees of freedom To make this work, we have to introduce a non-intuitive definition: I.e., the vector u and its scalar multiple are the same

Lines While we are at it, let us put the point into a vector, too: Which leads to the beautiful expression: Nonzero scalar multiple also does not change the point: So we introduce an analogous non-intuitive definition:

Example Ques: What does the vector [4, 6, 2] T represent? Ans: It depends. –If the vector is a 2D point, then the point is (4/2, 6/2) = (2, 3) -- divide by 3 rd coordinate –If the vector is a 2D line, then the line is 4x + 6y + 2 = 0, or 2x + 3y + 1 = 0 Points and lines are represented in the same way. Context determines which.

Lines Ques.: Is the point p on the line u? Ans: Check whether p T u = 0 Ques.: Which line passes through two points p 1 and p 2 ? Ans.: Compute u = p 1 x p 2 Ques.: Which points lies at the intersection of two lines? Ans.: Compute p = u 1 x u 2

Euclidean transformation 2D Euclidean transformation: is more conveniently represented as Again, we use 3 numbers to represent 2D point (These are homogeneous coordinates)

Perspective projection Nonlinear perspective projection can be replaced by linear equation where (x,y,w) T are homogeneous coordinates of (u,v,1) T :

Recap Homogeneous coordinates of 2D point (x,y) T are p=(wx,wy,w) T where w ≠ 0 We have seen three reasons for homogeneous coordinates: –simple representation of points and lines, no special cases –simple representation of Euclidean transformation –simple representation of perspective projection

Q & A Questions: –Is there a unifying theory to explain homogeneous coordinates? –How can they be extended to 3D? –Are they useful for anything else? Answers: –Projective geometry –Useful for planar warping, 3D reconstruction, image mosaicking, camera calibration, etc.

Euclidean  Projective Start with 2D Euclidean point (x,y) To convert to Projective, –Append 1 to the coordinates: p=(x,y,1) –Declare equivalence class: p=  p,  ≠0 To convert back to Euclidean, –Divide by last coordinate: (u, v, w)  (u/w, v/w) x=u/w, y=v/w

Ideal points What if last coordinate is zero? (u,v,0) Cannot divide by zero Projective plane contains more points than the Euclidean plane: –All Euclidean planes, plus –Points at infinity (a.k.a. ideal points) All ideal points lie on ideal line: (0, 0, 1)

Are ideal points special? In pure projective geometry, there is no distinction between real points and ideal points –Transformations will often convert one to another –We will freely make use of this, and often ignore the distinction However, distinction is necessary to convert back to Euclidean –Distinction will be made when we need to interpret results

Geometries Every geometry has –transformations –invariants

Stratification of geometries Euclidean similarity affine projective

Stratification EuclideanProjectiveAffineSimilarity ideal lineabsolute pointsone length allow scale allow parallel projection allow perspective projection

Cross ratio

Ray space

Unit hemisphere

Augmented affine plane ℓ∞ℓ∞ The line at infinity ℓ ∞ is “beyond infinity” ∞ ∞ ∞ ∞

Intersection of parallel lines ℓ∞ℓ∞ Note: Antipodal points are identified ∞ ∞ ∞ ∞ parallel lines y=mx+b (where m is the same) intersect at (1, m, 0)

Representing line at infinity ℓ∞ℓ∞ ∞ ∞ ∞ ∞ Cross product of two points at infinity yields ℓ ∞ = (0,0,1) (1,m 1,0) (1,m 2,0)

The strange world beyond infinity The line at infinity ℓ ∞ = (0,0,1) ax+by+c=0 This means 1 = 0 !

Line transformations If point transforms according to p’ = Ap How does line transform? u’ = A -T u

Conics Take picture of circle  ellipse No distinction between types of conic sections in projective geometry

3D Projective Points and planes Plucker coordinates for lines

Image formation 3D world point is (X,Y,Z,W) T 2D image point is (x,y,w) T Therefore, perspective projection is a 3x4 matrix P

Perspective projection Camera calibration matrix K

Homography Simple case is projection from plane to plane –Can be either world plane to image plane, or –image plane to another image plane, or –world plane to another world plane, –etc. 3x3 matrix is a projective transformation Called a homography

Euclidean homography Needs K

Essential and fundamental matrices

Relationship b/w FM and H Fundamental matrix and homography

How to compute homography Direct Linear Transform

Normalization Important