old stone barn - photogrammetry

During an all too brief stay back home, I was treated to some snow!

So I decided it was a great opportunity to see how well snowy things reconstruct!

I grew up in the Yorkshire Dales where there are lots of old stone buildings dating back several hundred years, some well maintained, some crumbling, but all with amazing character.

I found this one with good visibility all around so gave it a go.

I used a Canon EOS 600D with a vintage pentacon 28mm lens (also an experiment to test the suitability of older prime lenses for photogrammetry!

I used ISO 100, F5.6, I don't remember the exact shutter speed but it was fast enough not to over-expose the snow on the barn at any angle.

I took around 130 photos, wides all way around, a few close up and a few from the inside.

I'm using all open source software, VisualSFM for the point cloud reconstruction and MeshLab for surface and texture reconstruction.

The initial results look very promising! It failed to calibrate several of the internal photos, but that's not too surprising considering I was rushing at this point as it was starting to rain!

The dense point cloud came out really nice for the external surface, picking up some very pleasing detail, and reconstructing a lot of the snowy details.

The dense point cloud picked up a lot of the branches from the tree hiding inside the barn, so I cleaned up a lot of those points and a few other random bits before bringing into MeshLab.

This is the result of the Screened Poisson Surface Reconstruction, with Reconstruction Depth set to 10, not bad at all!

Obviously there's lots of missing points along the tops of the walls and the inside is a bit of a mess, but it's mainly the outside I'm interested in.

This is with the default vertex colors turned on.

After texture transfer (4k resolution)

I'm thinking to make this usable it might be worth adding a roof and door to hide the dodgy inner surfaces!