Measurement of PERCLOS using eigen-eyes

Monitoring of the level vigilance in humans through human computer interaction has been a field of interest for image processing researchers for long time. Numerous activities carried out by humans require constant vigil over a period of time. A lack of alertness can lead to precious human life and/or economic losses. A major cause of reduced level of vigilance is drowsiness. This paper presents a method based on Principle Component Analysis to classify eyes and calculate PERcentage eye CLOSure (PERCLOS) for drowsiness detection. PERCLOS is an established parameter to detect the level of drowsiness. Using Singular Value Decomposition eigen-eye spaces are created for fully open, partially open and fully closed eyes in the training phase from eye images. These eigen spaces are used to categorize test eye images to one of these three categories for calculating PERCLOS. Experimental results show nearly 98% accuracy for offline as well as for online categorization of eyes for calculating PERCLOS and level of drowsiness.