Inteligência artificial para a autenticação de condutores: uma abordagem utilizando redes neurais siamesas

• Ano de defesa: 2019
• Autor(a) principal: Souza, Andrey Gustavo de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Engenharia de Sistemas e Automação; UFLA; brasil; Departamento de Engenharia
• Palavras-chave em Português: Autenticação de condutores; Dados veiculares; Redes neurais artificiais; Redes siamesas; Identificação comportamental de condutores; Drivers’ authentication; Vehicle data; Artificial neural networks; Siamese networks; Drivers’ behavior identification; Engenharias
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/36883

The chronic problem of vehicle theft and robbery worldwide, and especially in Brazil, has grown considerably in recent years. In parallel with this problem, the increasingly abundant use of data has revolutionized various segments of the market through applications of computati- onal intelligence techniques for tasks previously difficult to solve using traditional algorithms. Aware of this reality, this work aims to develop a system based on an artificial intelligence model of driver authentication, which makes use of vehicle’s proprioceptive data, obtained th- rough the on-board diagnostics interface (OBDII) and inertial sensors present in smartphones. Different from other approaches that adopt this theme in the literature, the present work aims the authentication of drivers that were not used during the training step of the current model. For this, we used siamese neural networks for the driver’s authentication task to deal with this imposed limitation. Siamese neural networks are known for their performance in applications involving people identification, such as face recognition, even in situations where only few data are available for authentication. The adopted methodology exploits the ability of these networks to create embeddings from individuals’ data to carry out their later authentication through tech- niques based on distance, forming a decision function. It is also explored filtering techniques and features extraction, in this case, the use of sliding windows, which improves the perfor- mance of the siamese neural network. This combination of data processing and computational intelligence techniques has well performed the driver authentication task, even when the data have not been used for the Siamese neural network training. A ROC-AUC greater than 99 per- cent was obtained in real experiments, which indicates a good suitability of the siamese neural networks for the drivers’ authentication task.