Mining and comparative analysis of typical pre-crash scenarios from IGLAD

Scenario-based testing is crucial for considering the intended functional safety of automated driving vehicles. For the first time, pre-crash scenario mining research was conducted using worldwide accident data obtained from the Initiative for the Global Harmonization of Accident Data (IGLAD). First, data from the IGLAD database were analyzed and divided into four categories based on differences in traffic environments among countries and regions. Second, according to actual accident characteristics, fields and methods of clustering were selected, and 21 typical pre-crash scenarios were obtained using clustering and analysis. Finally, the typical scenarios were analyzed and compared in detail. Four conclusions were drawn as follows: 1. Considerable differences exist in traffic participant types, accident forms, and typical scenarios across countries and regions. 2. The third group of countries (3-G, represented by China and Brazil) in which accidents and pre-crash scenarios are the most representative and diverse is an ideal data source for the international scenario research. 3. The typical scenarios mined through clustering were highly consistent with the new test scenarios added in the Euro-NCAP 2025 Roadmap, but a few typical scenario elements which are critical for safety evaluations were still not covered in Roadmap. 4. Data from the IGLAD database still lacks a few important pieces of information for scenario research, such as obstruction of visual field due to obstacles, and the data representativeness need to be improved, therefore we recommend that IGLAD database adds some new data parameters to fit the further scenario research, and propose distribution requirements of accident data considering scenario elements. The analysis methods and conclusions presented used in this study could serve as guidelines or references for automated vehicle safety evaluations.