Driver Distraction Classification Using Deep Convolutional Autoencoder and Ensemble Learning

The study of real-time classification for driver distraction provides new insights into the understanding behavioral and cognitive reasons behind it. Among various approaches, deep learning models show better performance can be utilized a system. However, suffer from generalization performance. Ensemble learning, on other hand, combine numerous to improve generalization. Hence, ensemble model would advantageous overall classification. Approaches using are relatively scarce in this research field. This paper proposes an framework where novel dynamic is used classify based autoencoders set popular convolutional neural network (CNN) architectures. has also been tested two techniques, namely average-weighted-ensemble grid-search-ensemble, Our uses VGG-19 , xmlns:xlink="http://www.w3.org/1999/xlink">ResNet-50 xmlns:xlink="http://www.w3.org/1999/xlink">DenseNet-121 CNN with pre-trained weights xmlns:xlink="http://www.w3.org/1999/xlink">ImageNet dataset network. Hyperparameter tuning heads baseline performed get most optimum We open-source datasets, xmlns:xlink="http://www.w3.org/1999/xlink">State Farm Driver Distraction Dataset (SF3D) xmlns:xlink="http://www.w3.org/1999/xlink">Multimodal Multiview Multispectral Action (3MDAD), combined size over 80,000 images consisting 10 categories distractions that allowed increased reliability different situations lighting conditions. were fine-tuned transfer techniques. experimental results showed accuracy average weighted ensemble, grid search 88.91%, 89.04%, 89.13%, respectively, which higher than individual models.