Driver intent inference and risk assessment

Abstract

Both for driver assistance systems and highly automated driving, the in-depth understanding of traffic situations becomes more and more important. From the viewpoint of a warning driver assistance system, the authors analyze the requirements and challenges of risk assessment and driver intent inference in complex urban scenarios and provide a systematic overview of existing approaches. Furthermore, the ability of each approach to deal with more than two alternative maneuvers, partially observable feature sets, and potential interaction between traffic participants is evaluated. It is found that generative approaches and Bayesian networks in particular show great potential for driver intent inference, but it is also argued that more effort should be put into modeling the driver’”s situation awareness. Based on four concrete examples, the benefits of awareness-based situation analysis are demonstrated with respect to the avoidance of unnecessary warnings, the detection of occluded traffic participants, further improvement of the driver intent inference itself, as well as the prediction of the future trajectories of relevant traffic participants.