Fahrerabsichtserkennung und Risikobewertung für warnende Fahrerassistenzsysteme

Abstract

To avoid accidents, warning driver assistance systems require an on-line estimation of the current risk of collision. For that, a new method is proposed that – in principle – is able to deal with arbitrary traffic situations. This is achieved by the use of generative models to describe the expected driver behavior. Corresponding user studies in real traffic show promising results even when real time constraints are taken into account.