Training Intelligent Driver State Monitoring Systems: Design and Validation of an Experimental Procedure in a Driving Simulator Environment

Driver monitoring systems (DMS) have been developed to improve road safety by collecting data on driver status and behavior to profile their ability to drive safely. Nowadays, DMSs aim to monitor different types of drivers' emotions and distractions, allowing for the design of intelligent driver support systems that are aware of the drivers' state. To this purpose, they require carefully designed training datasets for supervised learning algorithms to ensure state detection reliability. In this paper we present an experimental procedure to collect training data for this kind of DMSs. The procedure was designed to induce different driver's state conditions in a driving simulation environment to the aim of capturing the related driver's data by means of the DMS sensors. We report the results of the implementation of the data collection procedure in an experimental study involving 28 subjects. Validation results in terms of participants' ex-post self judgements show that the procedure is effective to induce the expected drivers' state in terms of emotions, visual and cognitive distraction.