An Architecture for Situation-Aware Driver Assistance Systems

Current driver assistance systems merely use a minimum set of information. By using additional information of the environment hazardous situations can be detected earlier, more reliably and with a higher precision. This situational information has a significant impact not only on the hazard detection, but also on other modules such as the human-machine-interface or knowledge distribution between vehicles over suitable networks. For a number of reasons situational information is inherently subject to uncertainty. This may be handled by the use of dynamic Bayesian networks, which we suggest can be used to represent these dynamic complex systems. The objective of this paper is to provide an architecture for driver assistance systems that is aware of uncertain situational information in order to prevent accidents and reduce the number of traffic fatalities. It introduces the concepts of utility-based knowledge exchange based on the consideration of other partners' knowledge, the amount of inference that has taken place to identify a situation and current network conditions. Furthermore, we suggest a representation of hazard descriptions using Bayesian network fragments.