Search for transparent GSVideo. There have been some posts about this on the forum.
I agree with you, grafficjam24. Looking at the video it seems there is no real interactivity. The camera position is fixed. People are supposed to stand on a fixed spot which the animation is geared towards. The animation is just overlayed on top of the video feed (see for example 2:15). There has been some clever editing in the movie to 'cover up' some of the interaction problems and make it seem interactive. Not saying this diminishes in any way the merits of the project, just stating it as a neutral technical observation.
Programming-wise it seems to me it's basically two feeds/movies on top of each other (the last partly transparent or masked) and some video analysis to trigger the animation, right? The bigger value lies in the great 3D animations (from the correct perspective).