Towards threat metric evaluation in complex urban scenarios

Threat metrics are used to measure the criticality of traffic situations. Such metrics rely on models predicting future behavior of other traffic participants. These models need to be calculated with respect to lanes for handling oncoming traffic, but in urban scenarios, lanes are often not explicitly defined. In this paper, we present a first modeling and conformance testing approach for criticality metrics in urban scenarios. In addition, we analyze the influence of lane association on the quality of the models used for threat prediction. We evaluate our approach based on an urban intersection from the inD dataset. Our key finding is that such metrics are in general prone to both false positives and negatives if maps with static lanes are used.