Risk-based Driver Assistance for Approaching Intersections of Limited Visibility

This work addresses the general problem of risk evaluation in traffic scenarios for the case of limited observability of the scene due to a restricted sensory coverage. Here we especially concentrate on intersection scenarios, which are visually difficult to access. To distinguish the area of sight, we employ publicly available digital map data which includes, besides the general road geometry, information about buildings potentially blocking the driver's visibility. Based on the estimated area of sight, we augment the sensory perceived environment with potentially present, but not perceivable, critical scene entities. For those potentially present scene entities, we predict a, for the ego driver, worst-case-like behavior and evaluate the upcoming collision risk. This risk model can then be employed to enrich the traffic scene analysis with potentially upcoming hazards, which result from a restricted sensory coverage. Furthermore, it can be utilized to evaluate the driver's current behavior in terms of risk, warn the driver in case its current behavior is considered as critical and give suggestions on how to act in a risk-aversive way. By applying the resulting intersection warning system to real world scenarios, we could validate our approach. The proposed system's behavior reveals to be highly similar to the general behavior of a correctly acting human driver.