An extrapolation method on European accident data based on weighting and data harmonization

The validation of the safety performance of Advanced Driver Assistance Systems (ADAS) and highly automated driving functions (AD) is a main objective for their introduction. On this topic, a methodology is used to create simulation files of the pre-crash phase of accidents from police-recorded accident data: the resulting dataset includes various information (e.g., participant types, participant trajectories and speed profiles). These simulation files allow the reconstruction of the crash scene and pre-crash phase as well as the assessment of the effectiveness of ADAS. However, this dataset is based on police-recorded accidents from Saxony, Germany. Therefore, this paper focuses on developing an extrapolation method, in order to transform the database to the characteristic accident situation on a macroscopic scale. For instance, one aspect may be to assess the effectiveness of a newly-developed safety system at a European level, based on pre-crash simulation files. The methodology starts with a data review to link the simulation files with European accident data, then the extrapolation based on weighting factors is explained. It requires to find common variables between the two datasets, group the data by these variables and calculate the weighting factors. Due to the data difference and data categorization in countries’ accident statistics, the grouping and consequently a direct extrapolation are not possible: an in-between database must be created, which contains harmonized data of police-recorded accidents. This allows to group the data with homogenous variables and especially identical accident constellation categories. In addition, the process enables to extrapolate a small dataset of police-recorded accidents to the European level, or certain countries. It provides a method for the calculation of weighting factors, by defining reproducible requirements for the input data and for the variable groups to determine the weighting factors between each dataset.