Car driver skills assessment based on driving postures recognition

Recent developments of Human-Computer Interaction (HCI) technologies allow for a natural integration of car drivers in their vehicle environments. Using a driving assisting system based on HCI interfaces novice drivers can interact naturally in various driving situations and they can develop skills and abilities. One of the novelties of our on-going driving assistant project is represented by a multi-tracking HCI system based on a set of markerless sensors. In this paper we present how the proposed system enables us to recognize novice driver postures in various driving situations. Unlike previous approaches we do not consider body postures alone, but in relation with the following cues: eyes and head dynamics and contextual environmental information, i.e., states of the car and traffic information about driving situations. We built a rule-based expert system for driving posture detection that uses inputs from the data fusion of sensors. By analyzing the sequence of postures we can assess if the novice drivers maneuvers are performed in a correct manner and order, in relation to the car driving skills. The feasibility of the proposed system is verified through experiments. Results show that driver postures are recognized with an average of 96.43% accuracy and that car driver skills are enhanced using the proposed system.