EWMA based classification of driver state

Abstract

Most new methods for safety improvement rely on examination of the vehicle data and monitoring of the driver behaviour. The vehicle data may include steering wheel angle, the brake and gas pedal positions, gear, velocity etc. Driver physiological parameters are acquired using heart rate sensors, electrocardiogram, electromyogram, electroencephalogram, head/eye monitoring and tracking systems. Given a stream of input data the safety system should be able to determine the driver state in real-time. In this paper we use exponentially weighted moving averages for transformation of input data into feature vectors used for classification of driver state and investigate accuracy of this approach for datasets collected in driving simulator.