Driver support system by the prediction of driving maneuver

Abstract

This paper proposed an approach to design a driver support system. The driver support system which prevents unsafe confirmation is useful for reducing the traffic accidents during the intersection. The support system consists of the following three parts. The role of first part is to judge if the driver's confirmation is unsafely. The judgment is based on driver's gaze angles which are measured by CCD camera. The second part is to predict the driving maneuver in order to support in the early stage. The driving maneuver is predicted by Bayesian network (BN) which the parameters are learned every individual driver. Hence the prediction is adapted to each driver. The third part is to judge whether the support is necessary based on the output of previous two parts. Because of considering both confirmation behavior and driving maneuver, it is possible to judge more correctly. The parameters to design the system is tuned by and learned from data of driving simulator (DS) experiments, where the subjects drive repeatedly on the same road including some intersections with stop signs. Using these data, the system which is based on usual driving behavior is constructed. Finally, the effect of the supporting system is shown by the results of DS experiments.