The raw output from the Kinect has been reverse-engineered, but the processing software has not. So it's not very difficult at all to get the depth image, but don't expect to get limb positions.

Quote:Original post by Storyyeller
Do you think that finding a brightly colored object and trying to track that instead would work?

I guess you'd still have to find a way to filter out anything in the background which happens to have the same color.

Filtering out the background is easy. It's the region not moving. That is if you are holding an object and move it then those pixels will change (maybe using a threshold to correct for errors). It's the same way most webcam games work.

You can use the same idea to detect a users basic hand motions. You find the moving object then find the average depth. Then define a depth closer to the camera and anything that passes that plane can be grouped together and assumed to be a hand. I imagine it still requires a lot of effort to get the kinks worked out.

Something to keep in mind is that all machine vision related topics are absurdly patented. The value of Kinect out-of-box does not lie in sensors, but in hundreds of patent licenses that go into processing what comes out of controllers.

This is part of the reason why AR and machine vision related stuff hasn't seen mainstream adoption. Anyone just glancing into that direction would get nuked by lawyers as soon as they made a dime.