Comparison of Different Classifiers to Detect Symptoms of Drowsiness before the Vehicle is in Motion Using a Heartbeat Pulse Bracelet

Abstract

The drowsiness causes an average of 328,000 crashes per year. Many technologies had been created to avoid this kind of accidents. However, the number of fatalities which involves drowsy drivers still present due to the complexity and ambiguity of this common symptom. Computer vision and some Artificial Intelligence algorithms were implemented to solve the problem while the vehicle is in motion. Many of those have not been implemented in commercial vehicles because the drowsy driver detection with real conditions is not 100% reliable. Adding to this, some other systems review the heart rate variability (HRV) while the person is driving and detects if the driver is falling asleep. However, if the person doesn't use the sensor during driving, this method is not useful. This paper presents a different approach on how drowsy driver problem can be solved before a person starts driving. A heartbeat pulse had been used to measure and calculate the number of hours that a person sleeps per day, the hour of the day and quality of sleep. Then, data were analyzed using different classifiers to decide if the driver is a potential candidate to fall sleep on the road. As a result, a comparison between different classifiers showed that, for solving this particular problem, the best method is the Support Vector Machine classifier with an average of 91.23% of accuracy.