Drowsy Driver Detection System For Poor Light Conditions

Abstract

In recent years, the number of car accidents has been gradually increasing. A major factor in this increase is the drowsiness of drivers. There are many image recognition research studies on drowsiness detection but tend to neglect the case of low-light conditions. This paper focuses on detecting drowsy drivers in low-light conditions. In this paper, a behavioralbased drowsy driver detection system (DDDS) was proposed. The proposed DDDS applies a preprocessing algorithm that improves illumination and then focuses on detecting the driver’s face and eyes to calculate eye blink duration. A deep neural network model was used for facial recognition, and Haar Cascade classifiers were used for eyes detection. Every single eye was based to a Convolutional Neural Network (CNN) model to predict its state as either ‘‘Open’’ or ‘‘Closed’’. The proposed CNN was developed using a dataset of 1452 samples and gave an accuracy of 97.92% on the testing dataset, 15% of the 1452 samples. Different case scenarios in poor light conditions were implemented to evaluate the proposed DDDS. The system was able to identify the state of the driver even in harsh and severe poor light conditions.