Getting zero ROIs from stereoRectify function

asked 2018-07-26 04:44:40 -0500

kevinkayaks gravatar image

I ran cv2.stereoCalibrate and it returned error 0.97 pixels.

Now I am trying to rectify my images. cv2.stereoRectify seems very sensitive to alpha: many values of alpha return (0,0,0,0) in both returned ROIs. Some return nonzero values in one ROI. Others return nonzero values in both ROIs.

Is there any reason to this?

edit retag flag offensive close merge delete


and it returned error 0.97 pixels.

that's too large. try again with better images / coverage of the scene. err < 0.5 would be ok.

berak gravatar imageberak ( 2018-07-26 04:58:34 -0500 )edit

Any details on why stereoRectify fails with this larger error?

I have full coverage of the scene: I think this about the best I can do, unfortunately. The scene is underwater so the lighting isn't ideal.

kevinkayaks gravatar imagekevinkayaks ( 2018-07-26 11:55:50 -0500 )edit

@berak err = 0.526 after some improvements and I continue to get all black images from stereorectify. Any thoughts?

kevinkayaks gravatar imagekevinkayaks ( 2018-07-26 19:45:05 -0500 )edit

update: err = 0.320 after removing outliers. Fully black remapped images persist.

kevinkayaks gravatar imagekevinkayaks ( 2018-07-27 18:50:24 -0500 )edit

what methods did you use to reduce the reprojection erros?

rpat30 gravatar imagerpat30 ( 2019-04-06 02:26:08 -0500 )edit