2

What is the difference between using an A0 and an A4 camera calibration board ? Is there any ? If yes, what and why ?

For example, here they use A3 board http://se.mathworks.com/help/vision/ref/estimatecameraparameters.html

but here a much larger board : http://se.mathworks.com/help/vision/examples/stereo-calibration-and-scene-reconstruction.html .

Why not simply always use a small A4 board ? Why is such a large board necessary ?

Dima
  • 38,860
  • 14
  • 75
  • 115
jhegedus
  • 20,244
  • 16
  • 99
  • 167

1 Answers1

2

The size of the board you should use depends on the distance to the camera. Ideally, you want to place the checkerboard at the same distance from the camera at which you want to do your measurements. At the same time you want to have enough data points to cover your field of view.

If your checkerboard is too small, then you have two problems. First, at a large distance, you may not even be able to resolve the corners of the squares. Second, even if you can resolve the corners, the board will only cover a small part of the field of view in each image, so you would need many more images.

So, if you want to measure things a meter or two away, then an A4 board is sufficient. But if you want to measure things 10 meters away, then you want to use a much larger board with larger squares. And if you want to measure things kilometers away, then you just have to use the largest checkerboard you can find. :)

Here's a more detialed answer.

Community
  • 1
  • 1
Dima
  • 38,860
  • 14
  • 75
  • 115
  • The main question is: why do I need to place the checkerboard to the distance where I want to make my measurements? As far as I know light does not bend much in Earth's field of gravity... so if light rays are straight, what difference does it make how far the checkerboard is ? – jhegedus Mar 14 '15 at 12:53
  • Good one about gravity, but that is not the problem. The problem is that the camera model is imperfect, point detection is imperfect, and optimization is imperfect. – Dima Mar 14 '15 at 15:43
  • 1
    Here's an analogy: suppose you need to take some measurements (points) in the (x, y) plane, that you want to model with a straight line. You anticipate to use the model for x ranging, say, between 10 and 100. Your measurements will have noise, so you know in advance that you'll need to take more than the minimum (2). Now, ask yourself, which model would you trust more: one estimated on a cloud of measurements all clustered for x between 11 and 11.5, or one from measurements well spread along the whole (10, 100) range? Obviously the second. – Francesco Callari Mar 14 '15 at 20:28