Forum

Author Topic: Circular Fisheye Workflow...  (Read 2539 times)

TheAntiTheo

  • Newbie
  • *
  • Posts: 6
    • View Profile
Circular Fisheye Workflow...
« on: November 22, 2020, 09:41:56 PM »
I have been experimenting with a dual 190degree back to back  circular fisheye setup, I split, crop, mask and station the pairs of images before proceeding in metashape.

Given that metashape does feature detection and matching before lens optimisation, is there an advantage to me projecting the fisheyes to a more rectilinear format first to that features towards the edges of the FOV are less distorted?

I guess dense processing is ok as the images are projected first...

Thoughts?

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 14813
    • View Profile
Re: Circular Fisheye Workflow...
« Reply #1 on: November 23, 2020, 04:24:30 PM »
Hello TheAntiTheo,

I recommend to check at first, if the data can be properly aligned using Fisheye camera type selected in the Tools Menu -> Camera Calibration dialog.

If the images have some blank (black) areas by the corners or sides, it is also reasonable to apply the same mask, covering the no-data areas, to all the images and use Apply Masks to Key Points option in the Align Photos dialog.

Additionally, if the camera sensor locations are fixed relatively to each other, you can try using mutli-camera system approach (providing that you are using Professional edition) instead of Camera Station instances.

And can you please clarify, what do you mean, when you say "crop" in the pre-processing workflow description? Actually, applying any geometrical transformations to the original images, including rotation and cropping is not recommended, as it would affect the camera alignment step and proper estimation of the interior orientation parameters.
Best regards,
Alexey Pasumansky,
Agisoft LLC

TheAntiTheo

  • Newbie
  • *
  • Posts: 6
    • View Profile
Re: Circular Fisheye Workflow...
« Reply #2 on: November 23, 2020, 06:03:02 PM »
Hi Alexey,

Thanks for the quick reply. I should say my name is Chris, nice to meet you!!

I have to crop / split the original images as they come as two side by side circular panorama in a single RAW file. During the split i remove the unused black space at the edges and mask out the circle.

My tests so far using a left + right hemisphere split (only standard version for me) were successful, however I try to keep the yaw / bearing / orientation consistent between images as I move.

 I assume that if I were to rotate 90 degrees between image pairs (so that a feature that was in the centre of the fisheye is now at the edge) then association would fail due to high distortion even though the actual  motion is small. If that is true I could try to unwrap the image pair into a rectilinear view first to help.

I have a background in robotics so could do unwrap in openCV, just don’t want to try if the idea is silly...

Thanks,
Chris

Kiesel

  • Sr. Member
  • ****
  • Posts: 332
    • View Profile
Re: Circular Fisheye Workflow...
« Reply #3 on: November 24, 2020, 12:47:18 PM »
Hallo Chris,

I have use one shot cameras. As Alexey said, best solution is to use spherical camera model (panorama) at a first step for alignment and then multi-camera-rig or a precalibrated multi-camera-rig, but you need the Professional version for that.

With the Standard version I get best results if I not only take one (double fisheye) photo but additional a second photo, rotated 90° to the first, per station. This way you have 4 fisheye photos per station with much more overlap between them then the only 20° you have with a single shot.

Best regards

Karsten