December 30, 2019

Photogrammetry

The latest photography niche I've been dipping my toes into is photogrammetry. I am particularly interested in combining the resulting models with 3d printing for casting and jewelry production. I have been focusing on skulls so far because they are pretty cool and a surprising number of people I know have small collections of them around.

This involves taking numerous photos of an object and feeding them into a special program, which studies the differences in overlapping areas between key parts of each photograph, and interpolates a 3d model based on those differences. It's almost like a panorama in reverse or, for those of you that recall my VFX career, a camera track where the end result is a model rather than a camera path.

Fortunately, we already own most of the kit for such experiments while a turntable and additional lighting were cheap and readily available. The internet is such a bountiful source of information that it was very easy to learn how other people were achieving this goal and copy what has worked for them. Below is a picture of my basis setup which is now 5 for 5 in terms of success.
After knocking out the greenscreen in each photo, 3d Zephyr does some funky analysis to produce a loose point cloud and then a detailed model. There are numerous options in the photogrammetry software space and I picked this program since it combined a internally generated masking (key for turntable techniques) with a reasonable price for a perpetual license (software is increasingly moving to a subscription model and I don't want to pay a monthly bill).

I have found a sweet spot of 144 pictures (4 sets of 36, 2 with the object right side up and 2 more flipped) to provide overlap without taxing my time or attention unduly. This also yields a suitable model when reproduced at a small or medium level (the detail would not be acceptable for 1:1 reproduction, but will look amazing at 1.5" or so.)

The picture below shows the model in 3d Zephyr with each of the original camera positions denoted in blue.  Note that, once the greenscreen is removed, a static camera taking photos of a turntable is comparable to perfectly moving the camera around the skull at an even distance. Handholding the camera would be a nightmare since I use a combination of low ISO and small aperture which requires long exposures to compensate. The skulls aren't going anywhere so there's no hurry.
 Finally, the model is exported as an STL into zbrush. There I patch any breaks or fix areas that are low on detail or high on noise. Since we plan to cast some of these in the future, I also ensure all areas have a minimum thickness, add loops or pins to attach the piece, and add my maker's mark
I'll share more on the casting process when we get to that stage. 3d renders are neat but physical metal is a whole other ballgame of awesome.