OCamCalib vs OpenCV's rational model

[Matias Nitsche]

Hello,
I would like to know what would be the difference in calibrating a wide-angle camera using OCamCalib vs using OpenCV's rationl distortion model. I understand that OCamCalib model somehow includes the non-linear aspect of the distortion inside the projection itself and I imagine OpenCV treats the distortion separately. I wonder which model would be more correct/precise.

In my case, I'm interested in extracting the bearing angles to features extracted in an image. Moreover, I would like to try to perform feature extraction and matching between wide-angle images and I'm not sure how should I treat the images (in terms of how to undistort them).

Best,
Matias

[Davide Scaramuzza]

The OpenCV only works for camera fields of view up to 100-110 deg. OCamCalib is definitely more accurate than the OpenCV model for more than 110 deg
Best
Davide

> --
> You received this message because you are subscribed to the Google Groups "OCamCalib Toolbox" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to ocamcalib-tool...@googlegroups.com.
> To post to this group, send email to ocamcali...@googlegroups.com.
> Visit this group at http://groups.google.com/group/ocamcalib-toolbox.
> To view this discussion on the web visit https://groups.google.com/d/msgid/ocamcalib-toolbox/3eacf346-1aac-443d-b09a-01807b3854d6%40googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[v01d]

Hello again,

I tried ocamcalib and it worked really well. As I mentioned, I'm using this for fisheye lens undistortion.

Since I would like to perform feature extraction I will be working on the perspective-mapped undistorted image directly. However, I would like to obtain bearing angles of detected features in this undistorted image. For this, I would need to find the equivalent instrinsic parameter matrix (which allows me to go from pixel coordinates to normalized ideal coordinates). 

How would this K matrix be defined? I imagine pricipal point would be (px,py)=(w/2,h/2). However, I'm not sure about focal length. I guess the scale_factor should be considered.

Thank you!

Matias

You received this message because you are subscribed to a topic in the Google Groups "OCamCalib Toolbox" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ocamcalib-toolbox/C1PmYOnYmXs/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ocamcalib-tool...@googlegroups.com.

To post to this group, send email to ocamcali...@googlegroups.com.
Visit this group at http://groups.google.com/group/ocamcalib-toolbox.

To view this discussion on the web visit https://groups.google.com/d/msgid/ocamcalib-toolbox/DC60D973-4695-41BB-BFA9-111434F8645E%40gmail.com.

[phreak...@gmail.com]

Alternatively, I guess I could go from the undistorted image to the spherical coordinates by using the pre-computed perspective map (which goes from undistorted to distorted) and from there go to world with cam2world.
However, it would be best if I could go from undistorted to spherical coordinates directly, hence the previous question.

Matias

[phreak...@gmail.com]

Hi,
I'm again using Ocam calib and I still wonder about my last questions.

Best,
Matias

[akmaral.mo...@nu.edu.kz]

How did you get K matrix with camera intrinsic parameters (focal length) from parameters returned by calibration toolbox?