So what was in the SDK that PrimeSense was packaging with their reference design before the launch of the Kinect?
- OpenNI requires a calibration pose, while the technique Microsoft uses allows players to walk in and out of the frame. Kinect would not have succeeded if people had to calibrate it all the time.
- Microsoft does predictive analysis so it's much better in non-optimal environments (e.g. a living room) or in cases where it loses track of a body part because it's behind something.
- Also. To train it. Microsoft generated and processed millions of depth images from mocap data taken of people of all sorts of different shapes in sizes in all sorts of environments. It's insane. And of course they needed new algorithms to manage that...
- Lastly - no audio/speech or 3D facial expression tracking in NITE