Driver drowsiness detection using Convolutional Neural Networks-inspired features and Principal component analysis with K-Nearest Neighbors

Drowsiness and fatigue are the major reasons for triggering serious and severe road crashes in Zimbabwe and the whole world at large. The developments in technology in recent years brought support and backing to drivers using intelligent automobile systems. Several research studies used datasets of a single driver for training and testing. Some as well used mainly day time images for training and testing the models. Therefore, fatigue and drowsiness is a key possible field of study to prevent numerous number of sleep induced road crashes. In this paper, two methods of feature extraction were proposed which are the MLP (Multilayer - Perceptron), the CNN (Convolutional Neural Network). The PCA (Principal Component Analysis) method was used for dimensionality reduction. Based on these methods, five classifiers where used to detect drowsiness on the driver. The five classifiers used where the LDA, XGBoost, LR, Decision Tree and the K-Nearest neighbors. Experiments were done in order to examine the capacity and usefulness of the approaches contrasted with other techniques. Experimental outcomes demonstrate that the feature extraction technique of CNN provided high accuracy on the five classifiers. The KNN was the average best classifier with a 100% accuracy. Experimental results indicated that the PCA improved the classifiers. This study delivers important and significant answers in practice to curb motor vehicle crashes due to drowsiness.