That demo is impressive.
but, there are things to understand,
the demo is using a form of fixed-camera augmented reality. You will notice the camera is on a tripod, and never physically moves. It will likly have some very accurate snesors in the tripod measuring the direction the camera is pointing. This sort of AR is *really* hard to setup, but always looks fantastic in action.
You will notice he said it was running on standard hardware. This pretty much makes it unlikly to be using any sort of vision based AR tracking, as the camera is clearly a high end DV cam or something similar. My experience is that that any sort of computer vision work has real problems if you go above VGA at 30fps.
Also, you will notice there is image occlusion in the video. I'm not 100% sure how he had his hands occluded (or even if they were), it may be a form of motion detection on the video stream compensated by the known rotation speed of the camera. There has been some research done in this area to compensate for motion blur, etc, caused by a moving or rotating camera, Although that stuff most definitly isn't real time. There is also a fair bit of stuff out there for tracking hands using computer vision. I would guess this is what is being used. There is very little skin colour showing otherwise, which would make the tracking pretty fast as large areas can be rejected very quickly. There were some odd things happening with the hummer early on, and the depth was a bit off (hence he misplaced it first time). Although then again he never opened his hand, so he could easily have been holding a magnetic tracker (in fact, this is by far the most likly case due to how accurate the rotational tracking was).
The rest of the depth occlusion was pretty simple. They would have mocked up a 3D version of the table, and (real) objects on the table. This will be rendered to depth only. You can see this with the hummer, at one point it drives into a grey cube at the edge of the tabel, only to become almost completly occluded (even though the object it 'drives into' is a complex shape).
The animation and everything else is obviously very cool, designed to show the tech off more than anything. Flashy graphics are a good way to hide the limitations of a system
Things like the video feed inside the helicopter has been done *a long* time ago. Heck, novel (?) did a system with multiple cameras, and 3D interpolating system so the person you saw standing on your table, you could actually look around them from different angles (*very* hard to describe, incredible cool effect).
The majority of AR systems, of course, are not fixed camera. They are generally moving camera, often head-mounted using HMDs.
An example would be this company:
http://www.augmentedmedia.net/
Watch the video on their front page. it's halarious. Just a
minor technicality they forgot.*
They are, from what we can tell, a bunch of ex-students who are trying to turn a research project into venture capital. The problem is (from what I can tell) they are basing it on the research-only augmented reality library 'AR toolkit'. Not exactly modern, the ARtk has been around forever it seems. Also, they kindly mention they are planning to license it for games... o dear. Wouldn't want to be their laywer.
My own experience has mostly been with the AR toolkit too. But I have used other AR libraries.
My first AR project was the 'AR volcano',
high-res pic,
official page (with video). This is more traditional AR.
My most recent one I can talk about is 'AR starwars', which is showing currently as part of the Boston museum of science's StarWars exhibit (very cool exhibit!!). It's using something different again, a 'virtual mirror'. -
tv report about the exhibit - I'm in there, it's pretty obvious.. I'm making a fool of myself
, -
image 1 of my boss, -
image 2 - This is a rare case of a fixed camera in AR.
Future for AR?
Don't discount it. I'm involved in a couple of ventures bringing this stuff to market
It will happen.
For example, few people realise the 'glove' from minority report was actually real tech. I've been demoed it. It's damn cool
Sure the hardware is worth 500k or so, but in 10 years... $500. easy.
* - where are the computers? or cameras? or displays? 
