Reliable Crash Prediction (SUP)
Ahead of an unavoidable collision of two or more traffic participants, the expected crash severity changes with every action taken by any of the crash opponents. The outcome can depend heavily on whether a driver decides to brake or steer and thereby might drastically change the velocity and the overall crash constellation under which the victims hit each other. Due to these uncertainties, a statistical approach is necessary to describe the likeliness of a certain crash severity. In the research project “Reliable Crash Prediction”, a simulation framework is being developed to automatically generate a comprehensive crash database. Additionally, pre-crash data from real driving experiments on the CARISSMA indoor and outdoor facilities as well as high precision FEM-simulations are being used to extend the database. Finally, machine learning is applied to the data in order to train a regression model. With this model, the crash severity distribution of a real pre-crash situation can be retrieved online and in real-time.