Report for the
Invited ITSC Session on Real Time Non-intrusive
Driver Status
Monitoring
October 6, 2004,
Washington DC
Qiang Ji
Rensselaer Polytechnic Institute
The meeting went well and it lasted for the whole morning. A total of 6 speakers presented their works. The topics focused on real time and non-intrusive detection of drowsy drivers and estimation of driver workload. Most methods presented deal with video based driver fatigue detection. Specifically, Wenstrand et al from Agilent presented a new optical method for reliable detection of eyes using NIR illumination from a single image frame, therefore avoiding artifacts resulted from computing the difference image from two consecutive frames. Grace et al from Attention Technology/CMU discussed the their stand-alone drowsy detector based on measuring PERCLOS. The detector has been validated in laboratory. A field validation of the detector is currently being done by Virginia Tech. Richard also discussed the human factor studies they conduced on the usability as well as effective intervention mechanisms. Ji et al from RPI discussed their real time video based method for fatigue monitoring based on combining multiple visual cues and on explicitly modeling uncertainties and dynamics associated with fatigue and fatigue parameters. Trivedi et al from UCSD discussed and demonstrated various computer vision techniques being developed in the Laboratory for Integrated Safety Automotive (LISA). They are developing techniques for monitoring and analyzing driver head movement, driver emotion, driver posture as well as driver surrounding views.
Besides vision-based method, Eskandarian et al from GWU presented their work on monitoring driver
fatigue based on analyzing driver steering behaviors. Their method has been validated by both a DOT study as well as
their own study. Both studies involved different levels of sleep deprivation of
human subjects. Finally, Zhang et al
from GM presented a machine learning method based on the decision tree for
estimating driver workload.
Due to time constraint, the session adjourned a few minutes after 12:00 pm, without conducting the proposed panel discussion. The session was well attended, attracting about 20 peoples.