Why hasn't anyone built a plugin that says "sit still, now move left, now move right" and then just remove all the pixels that didn't move? Seems like it would be easy and work as long as the camera doesn't move.
I think current algorithms do a version of that, but it has to be coupled together with person/outline recognition and such. Taking the background in just once doesn't work because the pixels still change over time (due to lighting and such).
What's really impressive to me is that it looks like web apps like Google Meet and Discord seem to do this in JavaScript on the fly.
Is this a harder problem than I think it is?