Author Topic: Bumpy Geometry / Holes in mesh  (Read 5452 times)


  • Newbie
  • *
  • Posts: 1
    • View Profile
Bumpy Geometry / Holes in mesh
« on: February 18, 2015, 10:19:07 PM »

we've tried to make a mesh out of a series of photos of a person, but we have a problem with a very rough/bumpy mesh and missing backside of the body.

Our setup is made up of four cameras aligned vertically which are rotating around a person and they automatically shoot circa 22 pictures per revolution. So a complete 360° is taking four or five seconds. Since we have a green screen studio I made masks out of the keyed images. The image size is 5616 x 3744 pixels.

Unfortunately the mesh looks very bumpy and the back is missing completely. I guess the reason for this is a poorly lit model and the slight movements of the model. We have one single light box which is located above the model. I tried to increase the brightness of the images but the result is the same.

The footage of the bottom camera which should capture the legs is ignored by Photoscan completely, is it because of almost no overlap with other cameras?

Since we have the demo version I cannot save the project, so I attached some screenshots which should describe the problem. Help would be appreciated. Is there a way to improve the mesh with the existing footage?

Thank you!


  • Sr. Member
  • ****
  • Posts: 477
    • View Profile
Re: Bumpy Geometry / Holes in mesh
« Reply #1 on: February 19, 2015, 05:30:37 PM »
yes mainly what you said... more light is really needed... :P

but even though you have more light... 5 sec. shooting is a lot for a living subject.. so you won't get a very good model.. and it might be pretty noisy..

my 0.2 cent

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14927
    • View Profile
Re: Bumpy Geometry / Holes in mesh
« Reply #2 on: February 19, 2015, 07:42:23 PM »
Hello a__x,

I also suggest to improve the lighting. Check that ISO is not inducing hard noise, check that the details are not blurred (probably, higher aperture values should be used, like f/8). Also I can suggest to use portrait camera orientation that will allow to use the frame space more effectively.

Probably cloths with more distinguishable texture pattern should be used, since monotonous surface may have problems with the feature point detection.
Best regards,
Alexey Pasumansky,
Agisoft LLC