Do any of us have 360 cameras like a flip 360 or GoPro?
I'm interested in capturing 360 video from the center of a conference room table and I would ultimately like to use face tracking to identify every face and then follow it with a square that's say three faces wide and two faces high and create sort of a zoom call view.
Until I can figure that out, I would be satisfied with just cutting it into two 180s and stacking them.
But a lot of the cameras I've looked at seem pretty dependent on Windows software to process the video, and I was wondering if that's really true or if there's reasonable useful ways to use these things in Linux.
Are you looking to do this live or in post?
You can get >180 degree C-mount lenses, and USB3 C-mount camera bases that appear as a UVC device.
Live would be fucking epic. I hadn't even considered that an option. I would be satisfied with post.
Probably also add automatic captioning based on facial recognition because I know I can do facial rec already with some script
On Tue, Jan 27, 2026, 17:15 Richard Allen rsaxvc@gmail.com wrote:
Are you looking to do this live or in post?
You can get >180 degree C-mount lenses, and USB3 C-mount camera bases that appear as a UVC device. _______________________________________________ KCLUG mailing list -- kclug@kclug.org To unsubscribe send an email to kclug-leave@kclug.org https://kclug.org/mailman3/postorius/lists/kclug.kclug.org/
I have done things like this in the past. With OpenCV, I would probably start with a haar cascade face detector, then a tracking loop to filter the angle to center of face, then construct a region around the face, then pull that out using CV::remap to generate the virtual camera.
One challenge you might face is that the remap step is effectively digital zoom, and 4K@180° may not be a lot of pixels per face, depending on how far away people sit. Some systems merge many image sensors to provide good resolution of each face, and some use a cone mirror to provide a cylindrical FOV.
I agree the idea is cool from a technological perspective. 🙂
It sounds like you intend to associate facial images with other personally identifiable information. Is that correct? What privacy policy applies?
On 1/27/26 19:44, Richard Allen wrote:
I have done things like this in the past. With OpenCV, I would probably start with a haar cascade face detector, then a tracking loop to filter the angle to center of face, then construct a region around the face, then pull that out using CV::remap to generate the virtual camera.
One challenge you might face is that the remap step is effectively digital zoom, and 4K@180° may not be a lot of pixels per face, depending on how far away people sit. Some systems merge many image sensors to provide good resolution of each face, and some use a cone mirror to provide a cylindrical FOV.
KCLUG mailing list -- kclug@kclug.org To unsubscribe send an email to kclug-leave@kclug.org https://kclug.org/mailman3/postorius/lists/kclug.kclug.org/
It sounds like you intend to associate facial images with other
personally identifiable information. Is that correct?
Not exactly? It depends on where you draw the system boundaries.
In the scope of the computer vision system I proposed earlier, that works with a single generic face detector that draws a rough box around each face. It doesn't know who each person is, nor anything about them.
Process inputs: video stream and camera parameters Process output: event sequence (face found, cropped face, face lost) stored in cropped video streams.
This phase isn't too different than having a human doing the cropping, except the computer doesn't know anyone, and a human might.
Associating faces or seats would probably be done somewhere outside of this processing pipeline though.
So far as hardware though, what would you recommend? Insta360? Go pro? It some thing else, if I want it to not be a headache in Linux.
On Tue, Jan 27, 2026, 19:45 Richard Allen rsaxvc@gmail.com wrote:
I have done things like this in the past. With OpenCV, I would probably start with a haar cascade face detector, then a tracking loop to filter the angle to center of face, then construct a region around the face, then pull that out using CV::remap to generate the virtual camera.
One challenge you might face is that the remap step is effectively digital zoom, and 4K@180° may not be a lot of pixels per face, depending on how far away people sit. Some systems merge many image sensors to provide good resolution of each face, and some use a cone mirror to provide a cylindrical FOV. _______________________________________________ KCLUG mailing list -- kclug@kclug.org To unsubscribe send an email to kclug-leave@kclug.org https://kclug.org/mailman3/postorius/lists/kclug.kclug.org/
So far as hardware though, what would you recommend? Insta360? Go pro?
I think either of those will serve near equally well, and will be much less effort than a live solution. Get one that does 8K(both new models do). Avoid the older 4k/5.6k stuff.
If you want to compare optics vs a cellphone camera, you can compare using pixels per degree as an estimate. Digital compression will vary of course.
-
Billy Croan -
Richard Allen -
Shawn C. Powell