Stereovision: cvFindStereoCorrespondenceGC doesn't work proper, Incorrectly calculated parameters

asked 2013-12-17 13:52:12 -0500

Adamuswro gravatar image

updated 2013-12-17 13:57:19 -0500

Hi everybody!

I'm working with stereovision task. When I compared two matching algorithms (Block Matching, Graph Cut) using sample stereo pairs like Tsukuba Tsukuba

the diffrences was huge.

Block Matching

Block Matching

Graph Cut

Graph Cut

I used to think that using Graph Cut is only proper solution. But when I tried on the real images (of course after cameras calibration and rectification images) results were a disaster.

Input images (horizontal lines show correctness of calibration). Rect Images Block Matching Block Matching Graph Cut Graph Cut

Why Graph Cut is so bad?

Secondly, I checked values measured in calibration process (camera distance of each other - baseline and focal lenght). According to Learning OpenCV Computer Vision with the OpenCV Library (page 435) Q matrix has got all of this parameters. When I calculated Q matrix using cvStereoRectify(), I received the f=782.901 (real value - 3,85mm), T=3,01 (real value - 6mm). Images of cheeboard used for calibration - 25. Any ideas what could be wrong? Sorry for my bad english.

edit retag flag offensive close merge delete