Deep ResNet 18 and enhanced firefly optimization algorithm for on-road vehicle driver drowsiness detection

Abstract

The driver drowsiness detection (DDD) technology is based on vehicle safety, and this system prevents many accidents and deaths that occur due to driver drowsiness. As a result, it is monitored and detected when vehicle drivers become drowsy. The DDD method, which is aided by AlexNet and deep learning models, has limitations such as vanishing gradients and overfitting issues as the depth of the model increases. The enhanced firefly optimisation algorithm has solved the problem of lower optimisation exploration. The National Tsing Hua University Driver Drowsiness Detection (NTHU-DDD) dataset’s input image contains individual groups of female and male drivers of various vehicles. The Min-max normalisation method is a general method for normalising data. The convolutional neural network (CNN) is used to extract features from input images and images classified by the neural network. ResNet 18 refers to the deepest of the convolutional neural network’s 18 layers. A network of pre-trained models can be used to classify the model classified by the 1000 image objects. The state-of-the-art Hierarchical Deep Drowsiness Detection (HDDD) model with Support Vector Machine (SVM) assistance has an effective high dimensional space. The CNN-EFF-ResNet 18 models have a high accuracy of 91.3%, while the HDDD method has a higher accuracy of 87.19% than the ensemble and Pyramid Multi-level Deep Belief (PMLDB) methods in DDD.