Transfer learning using convolutional neural networks for driver distraction recognition

• Autores: Bazan Yaranga, Cristopher, Sanchez, Zaid, Rodriguez, Ricardo
• Formato: artículo
• Fecha de Publicación: 2018
• Institución: Universidad Nacional de Ingeniería
• Repositorio: Revistas - Universidad Nacional de Ingeniería
• Lenguaje: español
• OAI Identifier: oai:oai:revistas.uni.edu.pe:article/549
• Enlace del recurso: https://revistas.uni.edu.pe/index.php/tecnia/article/view/549
• Nivel de acceso: acceso abierto
• Materia: Transferencia de Aprendizaje; Ingeniería de característica; Conductores distraídos; Redes Neuronales Convolucionales; Transfer learning; Feature Engineering; Distracted driver; Convolutional neural networks

In the present work it is proposed to identify if a person is distracted or not, when he is driving a vehicle. This can be achieved by classifying images of drivers to determine if they are available or distracted using convolutional neural networks (CNN) and tools to improve the algorithm, which are Learning Transfer and Characteristics Engineering. Kaggle competition images are used to perform the training, in which you can obtain more results and obtain more results. Later the red extractor of characteristics VGG16 was used, which is a pre-trained model, from which it is lowered in its last layers to reduce the overfit and adapt it to our algorithm. The results obtained in the classifier gave us a training efficiency and validation of 99.30% and 99.46% respectively.