I’ve known for a long time that photogrammetry works really well for tracks. Both because work done by others was great, and because my own attempts worked out beautifully (see here, scroll way down for a swan track model. Or here for a dog track).
Recently, I happened to have a Facebook conversation with Emory professor Tony Martin, a really nice guy who works on tracks of extant animals. Tony has recently published a cool book I can only recommend, Life Traces of the Georgia Coast, and runs a companion website to it here. I essentially became aware of his cool work through the companion site’s blog. On this blog you can find an excellent post on an alligator track getting erased not by wind and waves, but by sand fiddler crabs.
Tony posted a bunch of excellent photos of the track, and that gave me the idea to try if I could generate a nice model from his photos – photos that were not taken with photogrammetry in mind. Thus, I chatted him up on Facebook, and he promised to send me the files. Because he is quite busy, this didn’t happen for a while. Then, I got into a kind-of fix with one of my research projects, and desperately needed a 3D model of a big alligator track. Thus, I contacted Tony again, and he sent the files right away.
When I saw them – 39 in all – I didn’t have much hope of getting a good model. Tony had taken a few overview photos, and many detail ones, but as he didn’t plan on photogrammetry he took nearly all photos from one side. The best side to show up the depth of the track, as the shadow of the “wall” of each footprint is visible. If you take the same photos from the other side, you won’t get a good depth perception. Here’s one example:
So in all, there were 39 photos, but many showed only one set of manus and pes (hand and foot) prints, and there were near-identical shots of each track. To get a model of as much as possible of the trackway I eliminated all the near-duplicates (they only cause tiny inaccuracies in the model, but give no new information), and also cut out all the close-up shots of tracks much destroyed by the fiddler crabs. All in all I only used 11 photos! Here’s how they aligned in PhotoScan:
As you can see, there is little variation in the angles of the camera. Some lateral shift, which is good, so there is a slim chance the dense point cloud will come out OK. I ran a model with medium density, and got this:
Wow! Not too shabby! The manus print at the bottom right has an obvious big hole in it, and turning the background colour white showed up thousands of other holes, especially in the deep shady areas of the deep claw tracks. But still – this is a good model!
Thus, I now ran a model with ultra-high resolution.
OK, lots of editing needed to close the holes, but this is a model I can work with! Look how well even the tiny sand fiddler crab sand balls come out. I was surprised that photos theoretically so unsuitable turned out such a good model. Good work, Tony 🙂
What now remained to be done was meshing the point cloud in Geomagic and editing it. I’ve not completed that yet, but the pes track is done. Note the pesky raccoon that also left a print (between the outer two toes of the pes track)!
And with that I am off to the US, for some research and then the annual Society of Vertebrate Paleontology meeting, this time in L.A.