However, it looks clear to me that a collision detection circle is added to the background visualization, and that the videography is designed to synchronize with that circle. The presenter's head represents the collision zone, where the circle is positioned. It makes the presenters head a "bounce away" zone. It's video/canvas orchestration.
God I wish I could scrub the video, though.
I would like to say it looks very neat. Hoping the guys chime in on how it parallels or surpasses existing tools. The canvas + feature mashup does appear to have great potential. There is limited reason to believe canvas elements are handicap accessible, unless you invest a large amount of time in custom markup configurations which the framework doesn't appear to have addressed.
Edit: OK, eventually I saw some of the stuff he was talking about, but only around the edges of the video frame. It looks like the most of the demo part was hidden behind the video.
Further edit: After trying it in firefox, I now realize that the demo just completely doesn't work in Chrome on my machine. The video plays with a big green rectangular background that pretty much breaks the presentation.