@daveC
I mentioned the Martin Lighting director as I know the designers and have seen the guts of the machine. It's pretty close to what you want to do, and it also works by triangulating position from ultrasonic sensors, so it's perhaps a benchmark for the complexity of sensing someone in real space. To make the MLD they ( numerous engineers and programmers) invested large amounts of money and more importantly time... the result is very good, but it is highly complex in technical terms. They are also really fantastic and savvy engineers, so if there was a simple solution using multiple ultrasonic sensors, they would have come up with it!
Alternately, have you thought of simply using the Very Nervous System by David Rokeby? It allows detection of movement and gesture in real space, using a video camera as input. I believe there is a software version that runs on MAX.
D