Idea: make the camera tiny (smartphone-like) and put it on suction cup. To pick up a call you don't tap a button; you grab it and put it on the screen on the other person's forehead (like a Bindi). To hang up, you pick it up and put it on a holder on the side of the screen. Boom.

Privacy? Check. Camera faces you only when your'e using it. Positioning? Check. Camera is closer to other person's eyes. Intuitiveness? Check. Like picking up a phone.

Wireless charging, alarm if it gets far away, etc etc.

I love this idea! I'm sending it to some Cisco engineers I know.

Sweet! Wish I was in the hardware space myself hahah (:

Video calls from a conference room with a dedicated below-screen camera are far better than from a laptop, IME.

Any given person in the other room is generally small enough on the screen that them looking at the screen instead of the camera is fine. But I'm not sure if a more overhead angle would be as good.

I think it comes down to the difference between eye lines and distance more than anything like the camera's placement above or below the screen. With a laptop any location the difference between looking at screen and camera is pretty drastic because it's pretty close to the user. In a conference room people are generally much farther away and smaller so it's harder to see the difference between looking at the screen and the camera.

It seems like the camera is detachable and can be placed anywhere. Probably has a clip, that's why they didn't build it into the bezel.

Perhaps they can process the image; We already do it with document scans from off-angle.

Wouldn't this require repositioning the person's eyes in the video? Not saying that isn't doable..

I doubt they're doing that, but it can be done realtime. Here's an example that runs in your browser, just move your mouse around and her eyes track it. http://alteredqualia.com/xg/examples/gaze_dolores.html

[Edit: Disclaimer, I work at MS but not on Surface. I also don't know if there is post-processing applied to the video I linked, so maybe it can't be done realtime.]

Thank you for the link - that's pretty awesome.

No, users look at where the other person is on the screen; not at the camera. So reprojection can be done to a known point of view.

IIRC, hololens does some of the reprojection magic already.

It's hurting my brain to think about this. Surely if you aren't looking into the lens then your eyes will not appear as if they are looking at the person since the lens's perspective is the source for the projection on the other side.

Like, if I take a picture of myself with a camera and look directly into the lens then my eyes are pointing directly forward. If I now continue looking at that same point but move the camera up on a step-ladder, the image I am going to get would not have my eyes aligning directly into the camera. How do you get around that?

Yes, and there are already AI/Machine learning solutions for this problem. The paper detailing this must be a few years old already.