No announcement yet.

Static v live head scanning issues

  • Filter
  • Time
  • Show
Clear All
new posts

  • Static v live head scanning issues

    First post here
    I’m, trying to do some 3d portraiture from photographs of friends as the subject.
    Most of the photogrammetry of heads I see used static objects as subject and there must be good reason for that. I am having mixed results with different data sets of RAW files, around 50 per capture per person.
    I’m using a dslr, the images are sharp and in raw format but i’m seeing a lot of noise and distortion in the current batch of photogrammetry portraits i’m doing using the trial Zephyr. This is in counterpoint to the first portrait I did using Zephyr which worked out great! The mask tool is brilliant!
    Is it just really difficult to get good head captures from a single point camera?

    Any ideas or setting to try?

  • #2
    Hi BillyCosta - It is really difficult to get good head captures with a single camera.

    The main cause of the problems is that people move. Even slight movement can totally mess up reconstruction, and personally I no longer do scans of people's heads with a single camera setup because it's quite simply headache-inducing. Multi-camera setups are much much better for this kind of thing, and even then, they have their own issues to overcome.

    Some tips:

    - Make sure you have plenty of light. The sun is great for this. The middle of the day with some cloud overhead will help diffuse the bright sunlight and minimize shadows.
    - Shoot quickly. Minimizing the time between shots will help combat the involuntary subject movement.
    - Shoot manual. Minimizing camera setting variation between shots will help aide the reconstruction.
    - Consider supporting the head of your subject with a prop they can lean against

    If you share your dataset with us (through GoogleDrive/WeTransfer (or whichever filesharing service you prefer) I can give you more directed feedback.


    • #3
      Many thanks for the reply and tips.

      I see, the issue is that if the central subject moves the software finds it difficult to compensate and i'm guessing thats the same for all photogrammetry software. So really photogrammetry is best suited to staic images. Manbe I just got lucky with the earlier sitters. I'll see if I can find a multicamera rig locally.

      After a bit of post process modelling however the previous zephyr captures were certainly good enough to carve out using my 4th axis cnc. I have left the computer to crunch away overnight using the exaustive setting and I expect there will be something useable after cleanup and modelling

      Just on another note. My workstation is offline at the moment and I cannot use the 'check online for precomputed camera calibration' function when importing from a Canon 550D. Will this affect the outcome.

      I'll see if I can upload the origin files. (I take it you hide the URL)


      • #4
        One thing you might consider is putting colored dots on the face using make-up. This provides 3DZephyr with clearer reference points in between the pictures. Human faces are actually quite feature-sparse, particularly in younger individuals. Old farts like me, with wrinkled faces, reconstruct a lot easier . You might try taking a couple of additional pictures without the dots to texture the face naturally. Won't be perfect, though. Even with a multi-camera setup, the mesh quality is modest in the absence of illumination with a dot laser pattern or using a structured light projector. Combined however, the results can be breathtakingly accurate.


        • #5
          Yeh dots or locator marks would be a good idea. Actually the last two heads i've captured have 'character' White hair doesnt seem to scan well, but hair highlights do