3D Photogrammetry Reconstruction Example
Example scripts and data from my Bristech talk. Slides at https://iainwallace.co.uk/2018/04/01/bristech-talk-april-2018/
The below are tested on Ubuntu 16.04, but will also work in a Cygwin environment with Windows MVE binaries and appropriate changes to the scripts (just point them at the location of the binaries).
It works on a Mac too (tested on High Sierra version 10.13.3) if you follow these build instructions..
Clone and build MVE in this directory:
git clone https://github.com/simonfuhrmann/mve.git cd mve make -j
If you want to use UVME to visualise the steps then also:
cd mve cd apps/umve qmake -qt=qt5 make -j
See the MVE repo for the prerequisites for it.
I want to make a 3D model!
Put all your images in a folder, then run scripts 0-5 in turn with that name as an argument.
Provided as a test/example are my Engineshed photos. using that as an example:
You'll now have a
engineShed_scene folder, you can open this scene in UVME and see the photos. This is where the output goes.
You'll run the sfm, have sparse features and camera locations. Reload the scene and look at the scene inspector. Check "draw sfm points" - you might need to move the view.
You now have per-image depth maps.
You can load
pset-L2.ply in the scene viewer, this is your dense cloud.
Now you have
surface-L2.ply to load - the surface reconstruction.
Now you've got the final cleaned mesh -
surface-L2-clean.ply. It should look as below: