ADVERTISEMENT
See Inside Scientific American Mind Volume 25, Issue 2

Eye-Tracking Software That Detects Mental States

The technology could help keep drowsy drivers awake



JASON LEE

A car detects when a driver starts to nod off and gently pulls over. A tablet or laptop senses its user is confused and offers assistance. Such interventions seem futuristic, but in fact they may not require any technological breakthroughs: a recent study suggests that with the aid of a standard camera, a simple computer program can learn to read people's eye movements to determine what they are doing and perhaps how they are feeling.

Psychologists at the University of South Carolina were curious if a computer could figure out what a person was up to based on their eye movements. They first had 12 people engage in four tasks, including reading lines of text and searching photographs for a specific printed letter. Each person repeated the tasks 35 to 50 times while a camera recorded how their eyes moved. Using a subset of those data, the team trained a simple computer program, called a naive Bayes classifier, to identify which of the four tasks each person was doing. In the remaining trials, the classifier correctly determined which task the person was working on 75 percent of the time, well above the 25 percent expected by chance.

Because the computer program is based on a flexible algorithm that is simple but powerful, this set-up could most likely be used to identify emotions or mental states such as confusion or fatigue, the researchers suggest in the paper, which appeared in September 2013 in PLOS ONE. With only a brief training period, a car's onboard computer—existing models are more than powerful enough—could learn how a driver's gaze changed as he or she became more exhausted. Further work, the authors suggest, could lead to devices capable of identifying and aiding people in need of assistance in a variety of situations.

Scientific American MIND iPad

Give a Gift & Get a Gift - Free!

Give a 1 year subscription as low as $9.99

Subscribe Now >>

X

Email this Article

X