Driver drowsiness demo by Tobii

Driver drowsiness detection is an active area of R&D where eye tracking definitely has a role to play. Tobii just presented a demo video showcasing their system. This is the first public demonstration where we actually get to see their eye detection visualized. One can only hope that this is the beginning of a more transparent approach from a company that previously have avoided showing the core technology in action. Judging by the video the eye region detection appears to be very robust, the dual camera setup clearly aids in reducing the frequency of false detection. For the specific application of driver drowsiness the accuracy of estimated gaze point-of-regard is less important as the primary area of interest is the degree of eye-lid closure, which in the domain is called PERCLOS (percentage of eye closure). This is potentially easier to accomplish compared to full gaze-estimation but placing the equipment inside a vehicle poses additional challenges. The imaging conditions will vary more compared to a traditional computer-display based system, for example direct sunlight, wider angles and head rotation. For a consumer safety application robustness is key, going from 90% to 99.9% detection rates is a clear challenge for all the companies attempting to push this technology into the automotive market. The application is clearly worthwhile and the manufactures want the technology, but is it ready?

On a side note, the sunglasses demonstrated in this video are lightly tainted. I did some testing last night with a pair of dark (black) Rayban and two 850nm IR lightsources (2x20 LEDs) and the eyes were barely visible. Guess it depends on the type of glasses and the illumination..