%0 Conference Proceedings
%4 sid.inpe.br/sibgrapi/2022/09.26.20.57
%2 sid.inpe.br/sibgrapi/2022/09.26.20.57.05
%@doi 10.1109/SIBGRAPI55357.2022.9991761
%T Fine-grained cars recognition using deep convolutional neural networks
%D 2022
%A Oliveira, Franklin Lazaro Santos de,
%A Macena, Arianne Santos da,
%A Barbosa, Otávio Azevedo de Carvalho Kamel,
%A Souza, Wesley,
%A Freitas, Nicksson Ckayo Arrais de,
%A Vinuto, Tiago Da Silva,
%@affiliation Federal University of Pernambuco
%@affiliation Federal University of Pernambuco
%@affiliation Federal University of Pernambuco
%@affiliation Federal University of Pernambuco
%@affiliation SiDi
%@affiliation SiDi
%B Conference on Graphics, Patterns and Images, 35 (SIBGRAPI)
%C Natal, RN
%8 24-27 Oct. 2022
%S Proceedings
%K intelligent transportation systems, fine-grained classification, car recognition.
%X Population growth and the high concentration of vehicles on urban roads have been negatively impacting urban mobility and the global environment, since the primary transportation modes occupy a lot of space on the streets and are one of the main polluting gas emitters. In this context of inefficient urban mobility and unsustainability, the Intelligent Transportation Systems (ITS) aims to solve or minimize urban traffic issues. ITS are also widely used in applications focused on traffic safety, such as vehicle recognition related to a traffic or law violation. For this task, the fine-grained vehicle classification technique is used mainly by advances in computer vision and deep learning. However, identifying vehicles by the model can be a problem because the same vehicle can be easily misclassified when observed from different perspectives, with different colors, or by similar models. Knowing these inherent issues from vehicle recognition tasks, Deep Convolutional Neural Networks (DCNNs) are commonly used due to their ability to extract features from images. In that regard, the goal of this paper is to evaluate some state of art DCNNs architectures, conducting experiments with three different datasets to identify which architectures have the best performance metrics in the refined car classification task within ITS context.
%@language en
%3 oliveira-20.pdf