After a lot of work I’ve finally got a working prototype for automating 3d-reconstructions with photogrammetry. Obviously the hardware is not built professionally but by me, and I’m more of a theory/software person. So I’m not using an industrial computer vision camera but a camera for photographers (X-T2), the parts for moving the camera are 3d-printed and in retrospect I think that I could have built this a lot better. But anyway, those are potential improvements for the second prototype in the future.
One major downside of this machine is that, since it uses only photogrammetry, it suffers from the disadvantages of that method. Specifically it really struggles with homogeneous or reflective surfaces. So the second version of this machine will probably include a range of different methods. Most probably Photometric Stereo and others.
Leave a Reply