1
$\begingroup$

I'm looking for some library with a function to compute an essential matrix from line correspondences in multiple images. All the computer vision libraries I know use point correspondences to compute the essential matrix.

I've seen some papers about how to calculate E from line correspondences but implementing it myself would be a project on itself, so I prefer to try some existing implementation first. Does anybody know of a library with such an implementation available somewhere?

$\endgroup$
1
$\begingroup$

When I think of lines in correspondence, my mind immediately goes for Pluecker representations, which are very suitable for such estimation tasks. In fact this work exists for camera pose estimation from lines and might be a good starting point:

Přibyl, Bronislav, Pavel Zemčík, and Martin Čadík. "Camera Pose Estimation from Lines using Pl\" ucker Coordinates." arXiv preprint arXiv:1608.02824 (2016). (link to the project page)

As you mentioned, there are various resources, especially when your camera is calibrated (I assume this because you ask for the essential matrix). If you really like to have a quick and dirty solution, you could either use the endpoints of the line segments (if you have them) or the intersection points of the lines.

In fact this just reminded me of one nice work, which generates virtual points on cameras along the lines and uses them for pose computation:

Line-Sweep: Cross-Ratio for Wide-Baseline Matching and 3D Reconstruction S. Ramalingam, M. Antunes, D. Snow, G. H. Lee, S. Pillai Computer Vision and Pattern Recognition (CVPR), 2015 (link to the project page)

Let's keep here posted with updates.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.