Photogrammetry Failures part 2

Progress was made today. We scanned the bed again, this time using over 80 photos, with a lot more care going into lighting and setting up the room. We also used adobe camera raw to batch process images to improve their sharpness and to raise the shadows a little bit.

The results were far from perfect though, with a lot of blurry textures, and some areas lacking enough information to put a good scan together.

a9560274ce3669fcb9809058153a8e23ce60888106e4c505c485fdf11d3dfc92

We also tried our hand at static-camera scanning, using a turntable and a key. We got mixed results with this and will continue trying. below is one of our most interesting failures.

There is a huge amount of green spill, due to how close the green-screen is to the objects and lights, there was also a problem with image alignment, resulting in only one half of the N64 being scanned correctly.

11980f2c4ca830bbfd42a262ebda396c

We will try again tomorrow, using a white screen instead, a stand for objects, and a patterned cover for the rotary platform, which will hopefully aid the camera alignment process.

Photogrammetry Failures

We are working on an interactive experience about an isolated teenager, and are in the midst of building the bedroom. yesterday I conducted my first tests with photogrammetry, which you can see the results of below.

150f5abcf5f449b3d1cce2e8004c0799

For those of you curious to try photogrammetry yourself here is a brief rundown of what went wrong.

  • The first issue was that there simply were not enough photographs, we only took 19, and the more you take, the better your results will likely be. Next time we will aim for around 60.
  • The second issue was there were not many defining features to the bed, it was mostly repetitive patterns, this meant some photos did not align correctly. The lighting kit we are using came with a green screen, white screen and ┬áblack screen, so we will set those up around the room to provide some points of reference for the alignment process. Similarly, we will adorn the bed with sheets and a duvet (as we initially intended) and this should help us to create a better scan.
  • We did not have a good stand for the camera, and as such, had to improvise by balancing the stand on objects. We got a new stand, which has a tilt-able bracket. This should help.
  • Finally, we only have three lights, as a result the bed is always going to be lit more brightly from one side, we hope to get around this in game, as the bed is only going to be view able from one side anyway.

Time for round two, I’ll update the blog with more information later!