Sistema de aquisição e classificação de dados para reconhecimento da presença das mãos do motorista no volante por redes neurais artificiais

Abstract

Advanced driver assistance systems are advantageous. In Brazil, the number of indemnities paid by the government increased 41% in the last 10 years and, of 100,000 traffic-accident occurrences, about 90% are by fault of irresponsible drivers. Aimed at reducing these occurrences, this work presents the development of a wireless system to acquire analog, inertial and vehicular data for classifying the presence of driver hands on the steering wheel and to help existing systems that needs this information. Basic concepts about data acquisition by automotive communication networks, analog signals, like electromyography and electrocardiogram, and inertial signals, like gyroscope and accelerometers, were essential to the whole system deployment. The driver behavior classification depended of a Multilayer Perceptron Neural Network developed in MATLAB™, which was fundamental in the identification of samples in according to some scenarios related to driver behavior evaluated graphically. The data acquisition system was reliable after being validated by an oscilloscope and the classification system detects, with 97.14 % of accuracy, when the driver behaves thoughtless while guides a vehicle.