Extracting subframes from spherical frame

[jumpjack]

I have a 360° camera capable of shooting both images and video.
I would like to use it to build 3d model of the interior of a room.
To do this, I need multiple images of the room from multiple points of view.

I was thinking about a script which makes what follows, but don't know which tool to use to perform the needed projection:
- downsample the video from 25FPS to 5 FPS or 1 FPS
- extract from each spherical frame some "planar subframes" (12? 24? how many?), showing only a small area of the sphere (i.e. almost undistorted, or "planar")
- group the subframes in 4 groups, one per each wall of the room.

Once I've done this, it's just a matter of feeding Autodesk Remake (or similar SW) to get a 3d model of each wall.

I think I can do the first step by using Virtualdub capability of downsampling videos, but how to perform second step?

I googled these tools, which one could be suitable? I'm on Windows 10.


https://ggnome.com/pano2vr

http://www.panoscan.com/Tools.html

https://github.com/d3/d3-geo

http://proj4.org/apps/cs2cs.html

http://www.flexprojector.com/

D3-Geo looks very powerful, but quite hard to understand.

I can code in VBA and Javascript; I don't want to install and manage heavy C/C++ SDKs.

The third step is easy, it's just a matter of grouping the images by heading of the subframe.

[Redrobes]

Not sure if the link you had is already the one I was thinking of, but Helmut Dersch had some super support for doing fish eye processing.

https://www.ptgui.com/

Edit: is the GUI for: http://panotools.sourceforge.net/
Notably, this one: https://www.ptgui.com/man/projections.html (See Flat !)
So there is some kind of script based interface to get at this ability to do what you need.

ffmpeg can certainly do the first step too.

Also, I am not sure you need a fish eye lens to get the source images to make a 3D model. The software is known as photogrametry or structure from motion. My blog here goes through all the steps from taking a photo to 3D object. It was written a while ago and some other tools have come to light as well. Not sure which ones I mentioned but open sfm (https://github.com/mapillary/OpenSfM) is a good start.

A tricky topic for sure but if you persevere you can get some good results.

[johnvanvliet]

why not just one of the many Structure from motion tools to build a point cloud and then mesh it like a any professional will do

open sfm is a great tool

and so is OpenMVG
https://github.com/openMVG/openMVG

[jumpjack]

Not sure if the link you had is already the one I was thinking of, but Helmut Dersch had some super support for doing fish eye processing.

https://www.ptgui.com/

Edit: is the GUI for: http://panotools.sourceforge.net/
Notably, this one: https://www.ptgui.com/man/projections.html (See Flat !)
So there is some kind of script based interface to get at this ability to do what you need.

ffmpeg can certainly do the first step too.

Also, I am not sure you need a fish eye lens to get the source images to make a 3D model. The software is known as photogrametry or structure from motion. My blog here goes through all the steps from taking a photo to 3D object. It was written a while ago and some other tools have come to light as well. Not sure which ones I mentioned but open sfm (https://github.com/mapillary/OpenSfM) is a good start.

A tricky topic for sure but if you persevere you can get some good results.

I'll check PTGUI, but I think it is not free?
The reason for using a fisheye images is that it actually "corresponds" to taking a dozen of shots for each fisheye image, so saving much work "in the field" (but resulting in more work at home).

OpenSFM gives ugly results: in the demos I just see a bunch of points which does not even slightly resemble a 3d model. I had a quick view, I'll check deeper.

[jumpjack]

These are the software I am using to reconstruct 3d model from 2d images:
Autodesk remake
3df Zephyr

Results I got with trial version of Remake are amazing:
https://sketchfab.com/models/b7fc9ce...a6500f9890fbf5

(obtained with 60 photos; full version (paid) can use up to 250 photos per model!)

I have now to try the free version (trial expired... :-( ).

The very good thing of Remake is that you don't have to do anything: just shoot, upload images, and download results (processing is performed by Autodesk servers).
Processing COULD be also done offline... but you need at least 16 GB RAM and an Nvidia graphic board, else processing won't even start.
Maybe 16 GB will be the entry level on 2020 PCs...