1

Does anyone know of any software out there that can generate a 3D model from 360 images or a 360 video?

I have a 360 camera and can take photos or videos, and want to turn it into a 3D model so I can bring it into a software like Blender and render some objects into it. I would then like to specify a camera path to fly through and basically make a custom video going wherever I want in the model.

I've come across the likes of:

https://www.youtube.com/watch?v=3IAK93U2QUI

Though it seems limited to enclosed spaces and I want to do something outside.

OR if anyone could suggest a software that would allow me to take normal images/videos and convert to a 3D model that'd be great too.

Thanks!

1 Answer 1

1

Daft as it sounds a 360 degree camera does not provide the data in a single shot to create a 3-D model. This is because it only has a single view of any one object that is in sight. Our eyes (for most people) provide a stereo view which gives us some 3-D information on close objects and our brains and experience let us make some assumptions about further away objects (but can easily be fooled).

To create a 3-D model you need lots of shots from many angles, ideally with either known camera positions or with enough overlap that this can be calculated in a process called photogrammetry. This normally consists of a Workflow with several steps some of which may require manual intervention rather than a single step.

  1. Feature Extraction (find things may appear more than once).
  2. Image Matching (find images that contain common features).
  3. Camera Position calculation (for each shot where was the camera and how was it oriented.
  4. Point Cloud construction
  5. Mesh Construction
  6. Texture extraction, scaling and application

One good free and Open Source set of tools & workflow is available from AliceVision.

4
  • Are your 6 points (fotogrammetry) only applicable starting from a set of 2D images taken from a moving camera, or also from a set of 360 images with moving camera?
    – Millemila
    Commented Feb 23, 2021 at 18:49
  • @Millemila a 360 degree image is just an extreme case of a wide angle and a video from a moving camera is just a large set of images taken a short time apart so this makes no difference. I would recommend using a) only keyframes & b) frames a reasonable time (and hence position) apart as the processing time grows with the number of images. Commented Feb 23, 2021 at 19:39
  • It is apparently possible to reconstruct a 3D model from only a single photograph, using a method called "approximate differentiable one-pixel point rendering." Commented Nov 22, 2021 at 18:07
  • @AndersonGreen: The linked video clearly states "Only a Handful of photographs" and "errors and flickering will occur where there is missing information". It is better at filling in the missing information than many other methods. Nice video though thank-you for the link. Commented Nov 23, 2021 at 6:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.