That said, I have yet to see a convincing demo of the new Kinect, so far only having seen increased resolution on the depth visualisation. By the very nature of the technology it will still have the same limitations as its predecessor.
The technology in the PrioVR may be a more cumbersome than ideal, but the proposed sensors are already smaller than those shown in the visualisation (http://i.imgur.com/oSBEloE.png - the middle ones are the new sensors). As for further miniaturisation, well it has to start somewhere (hence kickstarter, i guess).
As for power usage, this kind of device (low powered sensors with a single wireless hub) combined with a mobile phone and some kind of communication bridge would likely give you hours of full-system use (the PrioVR, this: http://www.kickstarter.com/projects/2041280918/vrase-the-sma... for display and some bridge to access the mocap data on the phone) off a simple 5Ah battery. Or you could go the oculus route, and just have a laptop in a backpack, powering everything through USB.
This isn't just about gaming. This is capturing full body movement for usage in a huge variety of applications. I'm worried that given the current traction this project has, it will fail to fund, and we're going to end up setting back immersive AR/VR for another decade.
I doubt that you could get hours of gameplay. Maybe if you were playing something very simple, you could do hours, but a fully immersive world would destroy battery life.
As for power usage, graphically intense 3d games on the iPhone 5 have about 2-3 hours battery life on a 1.5Ah battery (thats screen, CPU, GPU, etc), so I'd say its still perfectly realistic to expect hours. I don't know the power usage of this kit though, but i would expect it to be low solely based on its function.
For me, oculus aside, facial capture is probably the most exciting thing the new kinect can offer (due to increased depth and optical resolution), but it still has all the same limitations as the last one for full body capture (as its basically the same mechanism).
There are a number of full body mocap solutions built by third parties that utilize the data from multiple kinects. This certainly helps, but it still can't be as accurate as something like this, solely due to you being a coloured point cloud blob instead of collection of bones in the softwares eyes.
My full-body mocap needs are mostly sorted, but I'm extremely interested in facial mocap solutions right now because all the existing ones either exceed my budget or suck.
Brekel.com have a less expensive version, but I have no experience with it.
Open source alternatives seem thin on the ground, if not non existent.
Try faceshift, there's a free trial. It's mind blowingly awesome.
How does the setup work? I'm assuming you have to spend a while tying expressions to morph targets on the model?