Open AccessDriver Drowsiness Detection Based on Convolutional Neural Network Architecture Optimization Using Genetic Algorithm
Author(s) -
Yashar Jebraeily,
Yousef Sharafi,
Mohammad Teshnehlab
Publication year - 2024
Publication title -
ieee access
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.587
H-Index - 127
ISSN - 2169-3536
DOI - 10.1109/access.2024.3381999
Subject(s) - aerospace , bioengineering , communication, networking and broadcast technologies , components, circuits, devices and systems , computing and processing , engineered materials, dielectrics and plasmas , engineering profession , fields, waves and electromagnetics , general topics for engineers , geoscience , nuclear engineering , photonics and electrooptics , power, energy and industry applications , robotics and control systems , signal processing and analysis , transportation
In today’s analysis of traffic accident reports it becomes evident that most driving accidents result from driver drowsiness, fatigue, and lack of alertness. At these moments, drivers cannot react quickly to such changes in their state. Characteristics of drowsy driving include closed eyes, semi-open eyes, and yawing. In this research, a network architecture search mechanism has been used to obtain an optimal structure for a convolutional neural network that can effectively detect driver drowsiness. The contributions in this research include the following: Firstly, to enrich the dataset related to detecting driver drowsiness and address shortcomings in existing datasets, several videos showcasing various drowsy states have been extracted, and their frames have been labeled as either drowsy or natural. Secondly, an optimal convolutional neural network structure, including the number of layers, type of objective function, etc., has been obtained using a genetic algorithm, with FER-2013 used for achieving the optimal structure. Thirdly, transfer learning is employed, where the optimized network obtained from the genetic algorithm is considered a feature extraction component, and its fully connected layer is trained for the drowsy dataset. The proposed method outperforms other approaches in accuracy and precision, achieving an accuracy rate of approximately 99.8%.