Efficient PERCLOS and Gaze Measurement Methodologies to Estimate Driver Attention in Real Time

Drivers not being cautious enough is one of the major reasons for many of today's fatal road accidents. Drivers being fatigued or distracted have been identified as the main two reasons behind drivers losing their attention. PERCLOS and gaze estimation are two visual cue based parameters which can be used to estimate driver drowsiness and distraction respectively. This paper describes advanced and efficient methodologies for obtaining these two parameters. We use an infrared sensitive camera equipped with infrared LEDs in obtaining visual features of the driver. In PERCLOS estimation, for each frame, edge detected eye images are classified using a linear support vector machine. Exponentially smoothed vertical and horizontal movements of the pupils are taken into consideration in gaze estimation. The proposed methodology for eye state detection achieves real time recognition accuracy 83.64% whereas gaze estimation methodology achieves 80.5% accuracy.