Análise do comportamento do motorista baseado em aquisição de dados veiculares e instrumentação adicional do veículo

• Ano de defesa: 2021
• Autor(a) principal: Bonfati, Lucas Vacilotto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Ponta Grossa; Brasil; Programa de Pós-Graduação em Engenharia Elétrica; UTFPR
• Palavras-chave em Português: Motoristas; Avaliação do comportamento; Segurança no trânsito; Sistemas de reconhecimento de padrões; Eletromiografia; Motor vehicle drivers; Behavioral assessment; Traffic safety; Pattern recognition systems; Electromyography; CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA; Engenharia/Tecnologia/Gestão
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/27776

Today's cars have dozens of sensors to monitor and improve vehicle performance and communicate through vehicle networks (CAN) with various Eletronic Control Unit (ECU) that are responsible for decision making. Currently, the development of sensors to improve vehicle safety or customize the driver's experience is still growing. The objective of this work was to adapt a system to acquire data from vehicle sensors, from additional sensors installed on the vehicle and on the driver, aiming, with the aid of pattern recognition, to identify the driver and analyze his behavior. In this sense, many of the factory-installed sensors (such as speed, throttle position, steering wheel torque, braking monitoring, etc). However, some other sensors can be inserted in order to monitor variables that are not usually instrumented, such as the brake pedal, which was instrumented to analyze its dynamics and assess its importance in this driver behavior identification process. In order to validate and correlate the vehicle data, in the first phase, the driver's foot was instrumented with an inertial sensor and surface electromyographic signals were collected from the driver's leg. In the second phase, the steering wheel was instrumented to monitor its position and the inertial sensor fixed inside the vehicle (to simulate the yaw rate sensor). An experimental protocol was defined to seek to differentiate two driving patterns (here called calm and aggressive modes) and to verify if it is still possible to identify the driver. Three different classification tools were used – SVM, k-NN and Random Forests. Different combinations of parameters were tested in order to evaluate the performance of the classifiers based on accuracy and specificity values. It was found that there are superior results in the identity classification when driver data are included, as well as in the behavior classification. For the evaluation with in-vehicle sensors only, rates are higher when including brake pedal and steering wheel data compared to CAN-only data. As a result, when all available attributes obtained from the vehicle are combined, higher rates for Bagged Tree are obtained, with accuracy and specificity values, respectively, of 0.91 and 0.91 for behavior identification; and of 0.94 and 0 .74 for driver identification. When associated, initially with the classification of identity and, subsequently, of behavior, accuracy values greater than 0.97 were obtained.