Application of convolutional neural networks and digital image processing for eye status classification and drowsiness assessment.

Abstract

In recent years, the number of vehicles circulating on Brazilian avenues and highways has grown considerably. As a result, the time people spend driving their vehicles increased, which causes more stress, tiredness, and lack of attention. Due to these situations, the number of accidents has also expanded. In addition, driving requires a lot of attention and willingness. These facts were relevant to the growth in the number of accidents, which from 2016 to 2017 was 7,272, and approximately 38% of these were caused by sleepy drivers. In this work, the use of three Artificial Intelligence (AI) techniques will be highlighted for the development of the real-time application of the eye state classifier: Artificial Neural Network (RNA) and two Convolutional Neural Networks (CNN). These techniques were submitted to offline processing (which required a database with 811 photos) and online. The accuracy of the offline processes for the three techniques was approximately 77% for RNA and 95% for CNNs. The accuracy of the online tests for ANN, LeNet-5, and VGG16 were 57.48%, 90.52%, and 78.85%, respectively. The results of online tests showed that the most suitable technique for solving the proposed problem was LeNet-5.