Multimedia Fatigue Detection for Adaptive Infotainment User Interface

Abstract

Current vehicles are equipped with user interfaces that assist the drivers by presenting essential information such as navigation, speed limit, etc. In this work we present a fatigue detection model in order to build an adaptive user interface for vehicles that changes its properties according to the fatigue level of the driver. When the driver is fatigued, the interface parameters, such as intensity and color combination, are modulated to make the user interface more attentive and intrusive. At other times, when the driver is not fatigued, the interface properties are optimized for aesthetics and pleasure. In this way, the adaptive interface provides a warning to the driver while she is fatigued along with the routine essential information. We take a multimedia approach to measure fatigue by analysing the driver behaviour. The system captures driver behaviour with four media streams that capture: angular velocity of steering wheel, force on brake pedal, force on gas pedal, and grip force. These continuous media stream are fused together with other contextual parameters to detect fatigue. In the experiments, we evaluate two fusion techniques and 16 media stream combinations. It is found that the fatigue detection accuracy increases almost linearly with number of media streams fused. We also found that the steering wheel provides best cue of fatigue, while the gas pedal provides weakest cue. Personalized Bayesian Networks further enhance the accuracy.