Ask Your Question

# Calibration of a high resolution camera

Hi,

I'm trying to calibrate a high resolution camera of roughly 4000x3000 pixels hence 12 Mega pixels. (It's a canon EOS 1100D). I'm wondering what's the maximum calibration accuracy I can hope to reach with that kind of resolution. Here are some numbers I get:

-both with checker board pattern and asymmetric circle grid, I get a error around 1 pixel. I'm using between 46 and 60 valid shots of the board, with the camera mounted on a tripod, each shot carefully taken to make sure I have the best possible focus on the pattern. I cover most of the picture surface when combining all the shots. I used several grids from 6x9 points to 10x17 points. I varied all angles to make sure the algorithm converges. As 'good' calibration values should be between 0 and 1, I'm a bit concerned about a value around 1.

-on the other hand, if I reduce my high res shots to say 640x480, my error drops to 1/6.25=0.16 which seems to be more in line with some values mentioned on the web.

What's your experience? Somebody got to calibrate high res cameras with better results? Or is it really the best I can hope, given the noise/measures uncertainty and the algorithm used by Open CV?

Thank you for your help,

Pascal

edit retag close merge delete

## Comments

I have this questions too =(

( 2015-04-15 12:35:06 -0500 )edit

## 1 answer

Sort by ยป oldest newest most voted

I have this questions too :( . But i do a lot of experiments. I found a linear trend between the resolution and Repr.Err. Like you if i reduce the resolution, i have a smaller Repr.Err. I use a Canon EOS400 which full resolution is 3888x2592, i take 20-25 shoots changing inclinations of pattern ( i use a micro pattern 20x20mm more or less, symmetric and printed on glass, so the circles are perfect), with this settings i have a Repr.Err around 0,24. If i reduce the resolution to 972x648(4 times smaller) the error drops to 1/4 , approximately to 0,06, and if i reduce to 1944x1296( two times smaller) i have a Repr.Err = 0,12. I repeat this for a lot of set , i think that there is a correspondence . Maybe the explanation could be in the image with full resolution we have in the same space more pixel so the error is bigger, but i don't find any official document who explains this better.

more

Official site

GitHub

Wiki

Documentation

## Stats

Asked: 2013-01-02 09:27:53 -0500

Seen: 1,331 times

Last updated: May 20 '15