Tito’s Incblots


Tiki Photogrammetry / 3D Scanning

On the Carbide3D forum, I replied to someone who was asking about using a CNC mill to do 3D probing and then cut bas reliefs. I has seen mention of a technology called photogrammetry so I tossed that into the discussion as an alternative to physical probing. Then I got curious, so I got a copy of PhotoScan to learn about photogrammetry myself.

To experiment, I shot a little tiki statue that I have--being carved and wooden, it seemed like the kind of thing original poster might have been working on. I used just ordinary indirect daylight for lighting, at f16 and 1.5 second exposures (obviously with a tripod and remote shutter release). I was using a Nikon D200 with a 50 mm f1.4 lens.

I set my tiki on a Shimpo ceramics turntable under a sheet of white paper with a hole in it for the tiki, and another piece of white paper in the background. Like so:

And so:

I had the camera autofocus, then turned the focus to manual so the focus would be locked and thus consistent in all subsequent photos. I took 38 photos (tiff format) rotating the turntable a little bit at a time by hand.

Since then, I've built a stepper controller (for turning the item being photographed) with a shutter release relay (for triggering my iPhone camera) controlled by an Arduino. For iPhone triggering, I hacked an old set of apple earbuds with the in-line volume control. [photos and Arduino code coming later]


I then used Gimp (an open source cousin of Photoshop) to mask the photos, removing (almost) all but the tiki and saved the results as tiffs again:

I imported them into PhotoScan and after playing around with the settings ultimately ended up with an stl file and an obj file. The results I wanted took less than 2 hours of processing in PhotoScan on my bottom of the line mac mini--add that to the hour it took me to do the masking.

The model that PhotoScan made from the photos (and the point cloud derived therefrom) was 2.5 mm tall when I imported it into MeshCam. Unfortunately, MeshCam was able to scale it only about 3x. So I took the stl into Evolve (what I use for my 3D modeling of parts to be milled) and scaled it 64x so that it was about 6-7 inches tall (a little smaller than life size, which is a bit over 8 inches).

While I was in Evolve, I had some fun. I think maybe Dr. Jones would approve:

Or maybe more dramatic:


Finally, I imported again into MeshCam. I added tabs at the top and bottom, and used the following settings to generate gcode (using 1/8" square and 1/16" ball nose mills):

The predicted cutting time was something like 333 minutes. Here's the simulated first side:


And now coming full circle, here it is. It's in pine (easy to cut but poor at retaining detail), and by the time I worked out how I was going to secure the stock I had, I scaled it to about 5". Cut one side then flipped and cut the other.