Darren Stanley just completed his Master's thesis at RIT called "Measuring attention using the Microsoft Kinect"; the degree was in computer science. The research is actually similar to another recent study by Qiu and Helbig (2012), that used more complicated motion capture setups, and used some PEBL tests for validation.
The study used three of PEBL's attention/vigilance tasks, the TOAV, the pCPT, and the PPVT to measure and induce an attention task set on 20 participants. They simultaneously measured behavioral signatures related to posture and facial gestures, gaze, audio, and some other things. Then, they used regression analyses to determine which features predicted behavioral aspects of flagging attention (such as RT).
With such a large number of variables, many of which are correlated, analyses showed interesting but complex results. Eye gaze appears to be the most important predictor (we can tell if you aren't paying attention by where you are looking), but also variables such as head dip and body lean seem to be important, but there were also significant individual differences that should be recognized in future studies--different people may react differently to attentional lapses or sleep attacks. Obviously, more work is needed, but it is quite promising already.
This is some pretty interesting work that can hopefully be parlayed into tools that will help identify loss of attentional focus in interesting contexts, like for operators of complex systems, long-distance drivers, and the like.
Qiu, J., & Helbig, R. (2012). Body Posture as an Indicator of Workload in Mental Work. Human Factors: The Journal of the Human Factors and Ergonomics Society. doi:10.1177/0018720812437275 http://hfs.sagepub.com/content/54/4/626.short
Stanley, D. (2013). Measuring Attention using Microsoft Kinect. Masters Thesis, Rochester Institute of Technology. https://ritdml.rit.edu/bitstream/handle/1850/16642/DStanleyThesis5-10-2013.pdf?sequence=1