Abstract
An innovative embedded system has been developed by METRAVIB to evaluate in real time the driver workload for a better automotive safety and comfort. This system is able to recognise a certain driving situation from dynamic sensors information related to the car behaviour. This is a new product line globally aiming at "subjective" assessment of continuously evolving situations, from automatic systems supporting (or replacing) human experts assessment in order to estimate the driver workload from the driver's actions and his environnement.
It follows a recognition procedure finalised with a supervised learning phase which awards to the system all flexibility and opening. Dynamic data information (brake pedal, accelerator pedal, steering wheel angle, speed, indicator) are simultaneously collected on the vehicle at any time. Relevant parameters are then extracted taking into account memory effect of the dynamics. Then a decisional phase is applied to the extracted parameters in order to learn the decisional rule during the training phase and to apply this decisional rule during the application phase. The decisional phase is based on a hybrid approach that combines Neural Networks for stochastic information and Booleans rules for the certain information.
This research is based on several experimental campaigns in real driving situations. For several drivers, the performance of this system has been evaluated during a complete travel (with highway, country road and urban road) of one hour. Under these realistic conditions, the system provides 97% of good recognition for highway, 90% of good recognition for country road, and 72% of good recognition for urban situations. The system is now under tests on a VOLVO car and on a FIAT car for the development of an on-vehicle multimedia HMI (Human Machine Interface) able to harmonise the huge volume of messages oncoming from the new and traditional functions for the driving support. It now has the benefit to be connected to the CAN bus information. First system had been build in CEMVOCAS(**) project and improvements had been perform on COMUNICAR(*) project.