I came across Reactable a couple weeks ago, and have been playing with ReacTIVision, which is their computer vision engine. It seems pretty promising for quickly bootstrapping a table-based interface. (Although I do wish their calibration docs had a bit more detail.)