3D Modeling an archaeological site during excavation

I've been playing a bit with improving the photography during excavation at the site, including better and easier overhead photos (more on that soon). This season, when I was looking at a bunch of overheads I had taken, it occurred to me that I could likely generate a decent 3D version without too much more work using the AgiSoft PhotoScan software I had purchased a few months ago.

I threw the photos into the software and got the following (which I'm showing here as a screenshot). You can see big holes where the sides of the walls are not missing due to the overhead shots being fairly vertical and therefore not capturing the sides, but overall it looks pretty good and certainly made it clear that it was possible to generate good 3D models from photos taken in a few minutes on site. So the next time, I took not only a series of overheads but also a number of shots from ground level, circling the area I was photographing while snapping shots.

Here's what that approach yielded, here in a screenshot, but for the full effect, head over to the publicly available version at p3d.in, where you want the "burial" model. I've still got to work on optimizing the conditions in the modeler for getting good results: it's easy to generate something of the quality you see here, but getting better results takes a lot more cpu time and I haven't worked out a good workflow yet. Expect some more posts on this in future weeks.

What am I going to use these for? Well, first off they look pretty cool, which is something not to be neglected. It's a lot easier to capture people's imaginations when they can play with a little 3D version of a site. Second they of course form a more sophisticated documentation of the site for human viewers. Manipulating the model helps make clear in a way that 2D photos can't how various objects related to one another. Ideally the models will eventually be of good enough quality that we can do measurements on them. You can check out the meter stick in these images already (20cm to each red and white band). We try to capture relevant locations in our total station, but it's easy to miss something and the models mean we have less to worry about. It would also be great to incorporate the digital models into our GIS system. Right now they're already useful in reminding us just what everything looked like at a certain point in time, but it should be possible to make them part of our long-term digital database.

Other people are doing this sort of thing too. For example, check out William R. Caraher's blogpost on a talk back in April and his more recent one last week and another encouraging others to blog about their work (which is in part why I wrote this!). There's a little Twitter activity too, so go have a look at @BillCaraher @adamrabinowitz @sebhth @Pompeiana79 @adreinhard.