Driver Drowsiness Detection Based on Convolutional Neural Network Architecture Optimization Using Genetic Algorithm

Abstract

Drowsy driving is a major factor in many road accidents, which makes it essential to have dependable real-time detection systems to help keep roads safer. Detection of driver drowsiness presents a novel approach using convolutional neural network (CNN) optimized by a genetic algorithm (GA). The facial features of drivers are examined for the system to classify whether the driver is "Alert" or "Drowsy," thereby issuing warnings to prevent fatigue-related incidents. The Genetic Algorithm optimizes a few critical CNN hyperparameters dynamically, such as the number of layers, filter sizes, and dropout rates. This evolutionary optimization enhances classification accuracy and decreases overfitting in the model, thereby producing a much stronger and more generalizable solution. The CNN model was trained on a set of labeled facial images and tested for performance on a separate set for validity and applicability under real-world conditions. The achieved high accuracy with the optimized system is 91.8% and a billion low inference time of 50 milliseconds per frame suitable for real-time deployment with vehicles. This way, the driver monitoring system opens avenues for efficient and high performance through a smart marriage of deep learning and evolutionary algorithms. The results strongly suggest that the proposed method could be a promising option for enhancing Advanced Driver-Assistance System (ADAS) and thus building safer driving environments.