search

manhofer / Line3D

Line3D - Multi-View Stereo using Line Segments
GNU General Public License v2.0
37 stars 15 forks source link

can't find the paper that you mentioned #1

Closed lligen closed 9 years ago

lligen commented 9 years ago

hi, i can't find the paper that you mentioned, can you upload it to git repo?

manhofer commented 9 years ago

Hi

The paper will be presented at this years GCPR in October and is currently prepared for publication by Springer.

I can not upload it yet, but I hope soon. Please check our groups website http://aerial.icg.tugraz.at I will put it there as soon as possible

Best, Manuel Am 06.09.2015 15:00 schrieb "lligen" notifications@github.com:

hi, i can't find the paper that you mentioned, can you upload it to git repo?

— Reply to this email directly or view it on GitHub https://github.com/manhofer/Line3D/issues/1.

lligen commented 9 years ago

Well, thanks for your reply. By the way, is there any introduction or resource with respect to the Line3d project? Or, can this program afford real-time processing for robot pose localization? Any video available?

manhofer commented 9 years ago

Am 06.09.2015 um 15:48 schrieb lligen:

Well, thanks for your reply. By the way, is there any introduction or resource with respect to the Line3d project? Or, can this program afford real-time processing for robot pose localization? Any video available?

— Reply to this email directly or view it on GitHub https://github.com/manhofer/Line3D/issues/1#issuecomment-138086606.

Hi

The algorithm was developed in the course of my PhD program and within the founded projects PEGASUS and FreeLine, which focus on Power Pylon and Power Line inspection using UAVs. More infos here: https://aerial.icg.tugraz.at/funding.html

The whole thing (Line3D) is basically a multi-view stereo algorithm, similar to Furukawas PMVS but with lines. This means it needs the camera poses. So you need to perform Structure-from-motion first and then run Line3D as a post-processing. We have not yet tried to do a lot of real-time stuff, because pose estimation with lines is not so easy. Also, line matching without pose information is a bit of a pain in the ass. However, we have included similar methods to Line3D into an online SfM framework. That is not real-time but you can process an image per second or so. But the source-code for that is not (yet) available.

When you want to do pose localization with lines you should have a look at this paper: http://www.cv-foundation.org/openaccess/content_cvpr_2015/papers/Micusik_Descriptor_Free_Visual_2015_CVPR_paper.pdf

A video for our previous method (which is quite similar to Line3D but not open-source) is available here: https://aerial.icg.tugraz.at/videos/3DV2014_Hofer_(1080p).mp4

Best, Manuel

Dipl.-Ing. Manuel Hofer Schanzelgasse 24/23 A-8010 Graz

lligen commented 9 years ago

Thank you for your kind introduction. I'm so grateful.