Métodos de aprendizado de máquina para inventário de elementos de vias de trânsito

• Ano de defesa: 2021
• Autor(a) principal: Ribeiro, Johny Marques Borges
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciência da Computação
• Palavras-chave em Português: Machine Learning; Mapeamento móvel; Deep Learning; Lombadas; Sinais de Trânsito; Computação; Mobile mapping; Speed bumps; Traffic signals; CNPQ::CIENCIAS EXATAS E DA TERRA; Aprendizado do computador; Trânsito - Sinais e sinalização
• Link de acesso: https://repositorio.ufu.br/handle/123456789/31314; http://doi.org/10.14393/ufu.di.2021.110

Due to recent technological advancements, the cost of advanced driver assistance sys- tems (ADAS) technologies has decreased signiĄcantly, contributing to its widespread adoption in vehicles of all price ranges. In this reality, governments need to maintain traffic signs that compose the road network, since these signs are used as input to the ADAS. This work describes methods for both mapping speed bumps and creating road signs inventories as well. Machine learning techniques and convolutional network archi- tectures are compared for detection and classiĄcation of speed bumps and traffic signs respectively. Results of this work demonstrate the viability of the proposed method, reaching 96% of accuracy in the classiĄcation of speed bumps using Random Forest al- gorithm, and 94% of accuracy in the classiĄcation of traffic signs using Faster RCNN architecture.