From michael.rosen at gmail.com Wed Nov 11 14:09:33 2015 From: michael.rosen at gmail.com (Michael Rosen) Date: Wed, 11 Nov 2015 11:09:33 -0800 Subject: [Kwiver-users] Thoughts on support for dense reconstruction Message-ID: With an eye toward a a non-gpl tool chain to do orthomosaicing of overlapping aerial imagery, I'd like to ask about planned support for or suggested options for dense reconstruction. Given, say, a bundle.out file (or something equivalent generated by MapTK), how can we create a dense reconstruction like CMVS? Is this something that Kwiver / MapTK expects to support? msr -------------- next part -------------- An HTML attachment was scrubbed... URL: From matt.leotta at kitware.com Wed Nov 11 15:24:09 2015 From: matt.leotta at kitware.com (Matthew Leotta) Date: Wed, 11 Nov 2015 15:24:09 -0500 Subject: [Kwiver-users] Thoughts on support for dense reconstruction In-Reply-To: References: Message-ID: <834F8483-0C3D-4F79-8E07-2FD0EE84D7FE@kitware.com> Michael, By ?orthomosaicing? are you just looking for a 2D orthographic mosaic image or are you looking for dense 3D (or 2.5-D, like a DSM)? Either way the answer is "we have done that, but it needs to be better integrated?. Integrating these things is conceptually on our roadmap, but this integration work is not actively being funded by anyone at the moment, so it may take some time before it happens. There is code [1] in the VIBRANT project that will take a homography file produced by the MAP-Tk feature tracker or the VIBRANT stabilization process and produce a mosaic image. However, the mosaic is not necessarily in meaningful ground plane coordinates. Alternatively we can extract homographies from the cameras and point cloud produced by MAP-Tk [2] to get a sequence of homographies that mosaics images on the ground plane. On the other side, we have developed code [3] to take the camera calibration and point cloud (e.g. from MAP-Tk) and estimate a dense 2.5-D height map (aka DSM). This work was done in a separate experimental repository (not as well tested/maintained/documented) for an earlier project that pre-dates MAP-Tk. We would very much like to integrate these features within KWIVER more cleanly, but it will take a bit of work since each of these projects (VIBRANT, MAP-Tk, Super3D) came about more or less independently under different funding. If anyone is interested in volunteering to help or willing to fund this work, please let us know. ?Matt [1] https://github.com/Kitware/vibrant/blob/master/tools/warp_video.cxx [2] https://github.com/Kitware/maptk/blob/master/scripts/krtd_and_ply_to_homog.py [3] https://github.com/Kitware/super3d/tree/master/super3d/depth > On Nov 11, 2015, at 2:09 PM, Michael Rosen wrote: > > With an eye toward a a non-gpl tool chain to do orthomosaicing of overlapping aerial imagery, I'd like to ask about planned support for or suggested options for dense reconstruction. > > Given, say, a bundle.out file (or something equivalent generated by MapTK), how can we create a dense reconstruction like CMVS? Is this something that Kwiver / MapTK expects to support? > > msr > _______________________________________________ > Kwiver-users mailing list > Kwiver-users at public.kitware.com > http://public.kitware.com/mailman/listinfo/kwiver-users -------------- next part -------------- An HTML attachment was scrubbed... URL: