Eventos Anais de eventos
COBEM 2019
25th International Congress of Mechanical Engineering
Convolutional neural network approach applied to distracted driver classification
Submission Author:
Leandro dos Santos Coelho , PR , Brazil
Co-Authors:
Luiz Eduardo Soares Emidio da Silva, Rodrigo Negri de Azeredo, Luis Gustavo Tomal Ribas, Roberto Zanetti Freire, Viviana Mariani, Leandro dos Santos Coelho
Presenter: Roberto Zanetti Freire
doi://10.26678/ABCM.COBEM2019.COB2019-1039
Abstract
The number of passenger vehicle accidents has increased over the last decade, making thousands of victims worldwide. Among the different causes of accidents, it is possible to highlight problems originated by situations of distracted drivers. Therefore, a survey was carried out by the State Farm company for the production of images by on-board cameras in vehicles, in a controlled environment, with the objective of identifying risk situations. The situations could be recognized through pose estimation, evaluating hands movements and face position for the driver. The application of Convolutional Neural Network (CNN) is an efficient method in this type of classification task, since they already presented interesting results in the classification of images, showing as an efficient classifier option in terms of accuracy for this case. The obtained results showed a promising performance in terms of classification accuracy of the proposed CNN setup to the Kaggle’s State Farm distracted driver detection contest. The best results were obtained through a CNN with stochastic gradient descent (SGD) optimizer and ReLU (rectified linear unit) and an ELU (exponential linear unit) activation functions, with an accuracy of 98.91% and 98.89%, respectively.
Keywords
distracted driver, machine learning, classification, convolutional neural network, pose estimation
