Forum

Author Topic: Improved quality for spherical cameras by using their fisheye images  (Read 3434 times)

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Dear Agisoft development team,


Especially in very narrow spaces spherical cameras having advantages over frame or fisheye cameras. I have used them to measure vertical channels in old buildings and other narrow spaces. But when you work with spherical cameras, MetaShape have to handle them as ideal cameras, where no camera calibration can be done. Because of their construction they have parallax and therefore issues in stitch zone, which downgrades accuracy and quality of results you can get with them. Some of them save the original fisheye photos as double (or more) fisheye photo. It is possible to split them to single fisheye photos and use them in MetaShape. But because of very small overlap and the missing possibilityto define constrains between a pair of fisheyes you can't align them properly without of the help of the stitched spherical photos.

So my feature request would be:
1) to split to, handle and calibrate left and right fisheyes (in case of double lens spherical cameras) as single cameras groups with their own calibration parameters. (I do that by hand now)
2) For alignment it is needed to define 6, before calculated, constrains, 3 rotations and 3 translations of the second lens in respect to the first lens. This way it is possible to align extracted fisheye photos alone without of the help of the imperfect spherical photos and their stitching errors.

Would be really nice, thanks!

Best regards

Kiesel

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14965
    • View Profile
Hello Kiesel,

You can try to use the Multicamera System approach that is assumes to be used for fixed camera rigs. In this case you can allow Metashape to estimate the offsets between the cameras in the rig (both positions and rotational), while the calibration parameters would be estimated separately for each camera in the rig.
However, the requirement is to have some overlap between the images taken by the cameras in the rig.

I think we have discussed such approach when there was only a prototype of it in older PhotoScan version. But I can't find the image set that you have shared, so if you have a newer one, you can send it to support@agisoft.com or try to follow this approach by yourself (at the moment it is available in Professional edition only, though).
Best regards,
Alexey Pasumansky,
Agisoft LLC

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Hello Alexey,

thanks for answering that fast!

I'm surprised that you remember our discussion under all those others and answering those lot of questions (I have seen you have written over 10.000 answers now, wow congrats!

I will definitely have a look in the Multicamera System, we have the Pro version at our office now. So when I'm back at our office next week, I'll give it a try. The overlap between both fisheyes is very small, only 20 degrees of a sphere (both fisheyes have 190 degrees).

Thank you for the offer to send you an image set, I think I will need it.

Best regards

Kiesel

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Hello Alexey,

I have sent description, photos and two projects, one with spherical only and one with fisheyes only, to Agisoft support.

Best regards

Kiesel

PS: I think there is also some use for tunnels, caves, abandond mines and something like that. So a general implementation of such function, where you only input double fisheye images and perhaps some calibration data (or the possibility to calibrate double fisheyes with a round pattern) would improve Metashape a lot.


« Last Edit: May 23, 2019, 05:22:31 PM by Kiesel »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14965
    • View Profile
Hello Kiesel,

I've sent back to you the project generated using multi-camera system approach (each camera system contains two sensors). Master-to-slave offset has been estimated automatically during the alignment procedure.

Let me know, if the result is close to your needs.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Re: Improved quality for spherical cameras by using their fisheye images
« Reply #5 on: June 05, 2019, 10:01:03 AM »

Hello Alexey,

I have just seen your post. I had answered you already per E-Mail on monday.


Quote
Hello Alexey,




I have just seen your e-mail (wasn't in the office). Thank you very much!




Yes I'm interested how you have achieve this, because multicamera system approach is totally new to me.




But I have seen, that you have used rotated camera photos from respectively camera station. For each camera station, photos were made at 0°, ~60°, ~90° and ~120°, where 6 camera stations were made at a higher and 6 camera stations were on a lower position. This was only made for testing purpose, the idea was to use this for getting good camera calibration parameters from 2x6x4=48 photos and testing if this can improve camera alignment by more overlap, but normally we don't have this.




In a second step we want using those calibration parameters in a Metashape project where only  photos made at 0°are used, because it isn't possible to rotate camera in normal use, when camera is hanging in a vertical tunnel it is fixed and can only move down (and up), aditional it shouldn't spin to avoid blur introduced by low shutter speed. That's why we are interested in camera alignment from photos only made one at one position, in our case at 0°respectively 12 camera stations. Can you please try to achieve this also?



Best regards

Kiesel