Implementation of real time Visual Attention Monitoring algorithm of human drivers on an embedded platform

This paper presents an image based, non-intrusive, real time driver attention monitoring system to detect early symptoms of drowsiness. Driver inattentiveness has been identified as one of the principal causes of accidents on road. It is very difficult to monitor driver inattentiveness using physiological signals like heart rate, brain waves because of their intrusive nature. In this paper an image based non-intrusive method has been stated to detect driver inattentiveness in advance. Using Principal Component Analysis (PCA) face is detected in an image and then using PCA again, eye is detected from the face image. A comparison with Pattern/Template based method for eye detection has been presented. Once eye is detected the PCA based eye detection results are employed to categorize the eyes as “Attentive” or “Inattentive” based on weight vectors. Again using eye closure rating (PERCLOS) on this “inattentive” eye category inattentiveness is quantified and above a certain PERCLOS threshold an alarm sound is generated to indicate driver inattentiveness. This algorithm has been implemented on a stand-alone embedded development board, NI-CVS 1456, with an intel celeron 733 MHz processor and is found to run with an accuracy over 90%.