7
CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an “image of image” transform 1

CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

Embed Size (px)

Citation preview

Page 1: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

CS-498 Computer Vision

Week 7, Day 1 3-D Geometry

Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an “image of image” transform

1

Page 2: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

[Illustrate projection here]

2

Page 3: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

The equations for projection onto the image plane are:

i = x/z

j = y/z

These can be written as a homography…

1. Write a transform that maps x to i, j to y, and does not destroy z

2. Treat the result as a homographic point. (Note that the original wasn’t.)

3

Page 4: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

Transforms in 3D

4

Page 5: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

Rotation in 3D

This is a rotation around the z axis:

What axis is this a rotation around? In what direction?

5

Page 6: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

We can have homographic 3D points, too

Exercise:

Consider the equations

xnew = xold + tx

ynew = yold + ty

znew = zold + tz

Write the right-hand side of this equation as a matrix multiplication.

6

Page 7: CS-498 Computer Vision Week 7, Day 1 3-D Geometry Projection onto a camera image Projection as a matrix Rotation and Translation in 3D Homography as an

Homography as a “picture of a picture”

Suppose we take a picture of a picture.

The original picture is on a plane, and we can represent points on that plane as

7