Visual analytics using deep learning : drowsiness detection using deep learning

Abstract

The rising demand for remote working and learning makes the efficiency of undisciplined individuals an issue. Deep learning is a truly disruptive technology in the field of machine learning, and it is robust in data processing, especially data classification. This objective of this project is to build a Convolutional Neural Network for drowsiness detection features using deep learning technique. Transfer learning approach with pre-trained VGG16 model was adopted in this project to achieve this objective. Controlled training was conducted to find the optimal set of parameters setting values. The testing of different learning rates, mini-batch sizes, optimizers, regularizers were performed, and application of data augmentation. The resultant CNN is capable to detect drowsiness state images with an average accuracy of 68.46%.