Scribd
Upload
8 views2 pages

326 Intro

The document discusses stereo correspondence in computer vision, focusing on simplifying assumptions such as parallel image planes and horizontally aligned cameras. It emphasizes the relationship between two views and the concept of epipolar geometry, which simplifies the search for corresponding points. The speaker notes that these assumptions are not realistic for real stereo applications, which would require image rectification.

Uploaded by

saeb2saeb
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views2 pages

326 Intro

The document discusses stereo correspondence in computer vision, focusing on simplifying assumptions such as parallel image planes and horizontally aligned cameras. It emphasizes the relationship between two views and the concept of epipolar geometry, which simplifies the search for corresponding points. The speaker notes that these assumptions are not realistic for real stereo applications, which would require image rectification.

Uploaded by

saeb2saeb
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

1

00:00:00,140 --> 00:00:01,980


Welcome back to computer vision.

2
00:00:01,980 --> 00:00:05,370
Today we're going to talk about, stereo correspondence.

3
00:00:05,370 --> 00:00:09,410
Up until now we've defined the epipolar geometry that talks about how

4
00:00:09,410 --> 00:00:13,980
the two views relate, and how if you have a point in the left image,

5
00:00:13,980 --> 00:00:18,340
then if you know the camera's relation, it's a one dimensional search.

6
00:00:18,340 --> 00:00:20,640
And in general, the epipolar lines can be,

7
00:00:20,640 --> 00:00:25,580
well, arbitrary, not arbitrary, but skewed in a variety of ways and located.

8
00:00:25,580 --> 00:00:28,644
But today, in order to make life easier, we're going to assume a bunch of,

9
00:00:28,644 --> 00:00:31,100
simplifying assumptions.

10
00:00:31,100 --> 00:00:34,010
For example, we're going to assume, basically the geometry that we

11
00:00:34,010 --> 00:00:39,500
drew out last time, of parallel, or coplanar actually, image planes.

12
00:00:39,500 --> 00:00:43,290
We're going to assume the same focal lengths of the two cameras.

13
00:00:43,290 --> 00:00:44,280
We're going to assume that they're,

14
00:00:44,280 --> 00:00:47,480
the cameras are horizontally aligned at the same height.

15
00:00:47,480 --> 00:00:50,840
So we're, and that the images are pulled out such that the epipolar lines,
16
00:00:50,840 --> 00:00:57,030
are horizontal, and, that they are, in fact lined up, so a xy location

17
00:00:57,030 --> 00:01:01,380
in one image is at a different x but the same y location in the other image.

18
00:01:01,380 --> 00:01:04,620
So look, that's a lot of assuming, and for real stereo, you wouldn't be

19
00:01:04,620 --> 00:01:07,390
able to do that, you'd have to do a rectification of the image first.

20
00:01:07,390 --> 00:01:10,590
We'll talk about that, couple lectures from now.

21
00:01:10,590 --> 00:01:12,950
But for now, we'll make the assumptions which will allow us to

22
00:01:12,950 --> 00:01:15,510
attack the correspondence problem more easily, and in fact,

23
00:01:15,510 --> 00:01:15,510
it's the way you're going to be doing correspondence on the problems sets.

You might also like