Driver Drowsiness Detection System using Convolutional Neural Network

Abstract

Driver drowsiness is a condition that impairs judgment. Many accidents have occurred due to sleepy drivers behind the wheel. Multiple studies require drivers to wear health sensors to gather useful physiological health data. This technique is intrusive and makes the driver feel uneasy while driving. Also, this method is less successful since the sensors must remain in specific positions for reliable results. Accurate detection of tiredness and prevention of a potential mishap on the road is a tedious and practically impossible process. Hence this paper proposes to build a driver sleepiness detection system utilizing a behavioral approach to warn drivers before an accident occurs. This system can help reduce traffic accidents and save the lives of drivers. This work trains a Convolutional Neural Network and uses it to assess whether the driver's eyes are closed or open. The dataset comprises images acquired from a large portion of the MRL eye dataset. Before training, the proposed model processes the images in the dataset using computer vision techniques such as edge detection, grayscale conversion, and dilation. The Google MediaPipe Face mesh model is then used to track facial landmarks from video frames in real-time. The eyes region is retrieved, processed, and fed to the proposed trained model for prediction. The model detects drowsiness and alarms the driver to take safety measures. This paper proposes and implements a CNN model that achieves an overall accuracy of 95%, outperforming all previous studies on drowsiness detection.