For my dissertation this year (which I hand in next Friday! Time flies), I've been investigating the use of the Microsoft Kinect in the development of a gesture-based interface. I've built a program that uses the Kinect to find your hand, track it, and then recognise the gesture you perform.
Two words: Gorilla. Arm. I had to run a series of tests to see how it compares with other gesture-based inputs (programs using a webcam, for example), and five minutes into each test, my arm was killing me.
I think motion detection will work a lot better when we can make the gestures more general or when they are imitating the actual gesture you are doing. Like the shooting a gun, wtf.
<script type="text/javascript">alert("remember when scripting attacks worked?");</script>