The use of prospective effectiveness assessment based on simulations, for estimating the safety performance of Advanced Driver Assistance Systems (ADAS), has become highly relevant due to the high number of scenarios to be assessed, which puts limitations on physical testing. This study aims to set up a method that defines the baseline data used to assess ADAS in VRU related traffic scenarios, based on real world accident data. The method further allows the weighting of results to take the differences in the frequency of velocity values into account when calculating the accident avoidance potential. As input data, the German In-Depth Accident Study (GIDAS) database is used, filtering accidents only for the most common scenario types and relevant to the technology being assessed. For each scenario type, and for the Vehicle-under-Test (VuT) and VRU respectively, a histogram of velocity values with a fixed bin size is computed. Velocity bins are added iteratively to the velocity range until the desired minimum share of all accidents, or the respective scenario type is reached. An exemplary effectiveness study with velocity ranges defined by the method is conducted where four generic inner-urban pedestrian crossing scenarios are investigated, both with and without sight obstruction. The scenarios are simulated with and without an exemplary state-of-the-art autonomous emergency brake system (AEB) to allow a baseline-to-treatment comparison. The accident avoidance potential is computed, and results are weighted to real traffic accident frequencies. For each of the four investigated scenario types (crossing from left or right, with or without sight obstruction), velocity ranges for the pedestrian and VuT were defined by the proposed method, such that the simulated cases cover at least 80% of the velocities observed in accidents in real traffic. In addition, variation of impact parameters (e.g., impact location) led to further diversification of the scenarios. By weighting the results with the respective relative frequencies, it was determined that the AEB system can avoid 50% (pedestrian crossing from right, with sight obstruction, P-CRwSO) to 90% (pedestrian crossing from left, without sight obstruction, P-CLwoSO) of the frontal impacts in the investigated scenarios. Using an Injury Risk Function (IRF) for frontal impacts, it was determined that the average risk for KSI (killed or severely injured) injuries could be reduced by 93% (P-CLwoSO) in the best case and 56% (P-CRwSO) in the worst case. Only generic scenarios with straight paths for the VuT and VRU are considered, while real accidents might show curved paths. Furthermore, the distributions of VRU and VuT initial velocities are assumed independent. In addition, driver responses to emergency situations are not included in the simulations, despite some drivers in real accidents in the GIDAS showing a brake response. Driver emergency response models are not within the scope of the current study. The study focuses on a method with adjustable parameters (i.e., the proportion of velocities to be represented) to adapt to the desired proportion of cases/scenarios to be addressed. Uniform variation and weighting of results by velocity frequency offers an advantage in comparison to Monte Carlo methods, since it can be ensured that the whole range of possible values is covered in a specified granularity, requiring fewer simulations. The method provides a more controllable coverage of velocity values than direct sampling from accident databases, the cases are more diversified, and the implementation is less complex. Due to uniform sampling, the presented scenario generation method is simple to apply, offers a controllable coverage of velocity values and provides more scenarios than sampling from databases for virtual testing and effectiveness assessment of ADAS. The most important required input are velocity histograms for GIDAS accidents of the relevant scenario types. The method was demonstrated exemplarily for a pedestrian AEB, but can be applied in principle to a variety of other ADAS and road user types.

Dr. Harald Kolk, Senior researcher, Virtual Vehicle Research GmbH