Recitfy Images before triangulate points

Hello, I have a question regarding the new triangulate function. I have 2 almost parallel cameras and would like to run a triangulation of certain points in the image. Do i have to rectify these images before extracting the matching points or do I extract the points of the original image?

Thank you

Answer 1

[Joep]

The matlab function uses the cameramatrix and undistorted images. So you should not use both of the image. You should create first a undistorted image.

uImage1 = undistortImage(Image1,stereoParams.CameraParameters1);
uImage2 = undistortImage(Image2,stereoParams.CameraParameters2);
point3d = triangulate(pointImage1, pointImage2, stereoParams);

[hochers]

So before using the triangulate.m function i need to use rectifystereoimages.m from the computer vision toolbox?

[Joep]

Sorry i was wrong about that i changed my comment. A example is explained here: http://nl.mathworks.com/help/vision/ref/triangulate.html

[hochers]

Thank you! the triangulate function rectifies the images internally.

[Joep]

I guess so because you expect that you should rectify, to make the epipolar lines parrellel. So you can use triangulate but in example they don't rectify so I assume it is in the function it self indeed.

[Dima Lisin]

You should undistort the images before using triangulate, but you should not rectify them. The triangulate function uses the linear triangulation algorithm from Hartley and Zisserman, which does not require the images to be rectified.

[Federico Perrotta]

Hi Dima Lisin,

I' m very interested in your comment. Could you explain me how the triangulation is possible without a rectification procedure? Does it follow that every stereo configuration would be possible to reconstruct 3D points?