Uncertain inverse traffic accident reconstruction by combining the modified arbitrary orthogonal polynomial expansion...

Traffic accident reconstruction accomplished by traditional deterministic inverse method is highly user-interactive and time-consuming and its solution is hard to convince in court due to various uncertainty in measurements, calculation and modeling. To rapidly identify the unknown impact parameters and simultaneously quantify their uncertainty, an uncertain inverse traffic accident reconstruction is accomplished by combining the modified arbitrary orthogonal polynomial expansion (AOPE) which can accurately and effectively construct surrogate models and novel optimization technique which solves the uncertain inverse problem by deterministic method. First, a dynamic model is built to simulate a traffic accident. Then, using the optimal Latin hypercube design, the impact responses fluctuating greatly with the change of unknown impact parameters are selected as the evaluation index of reconstruction quality. Using the random forest algorithm, the impact parameter importance is evaluated according to the SHAP values. Subsequently, a modified AOPE is presented based on the impact parameter importance measures. The mappings for an actual traffic accident are approximated using the modified AOPE to reduce the computational burden and guarantee the fitting precision. Finally, based on the novel optimization technique, the uncertain inverse traffic accident reconstruction is defined to search the optimal distributions of unknown impact parameters which make the distributions of simulation results of impact responses match these of accident investigation. The uncertainty is propagated by the Monte Carlo simulation in the optimization process. An actual vehicle-to-two-wheeler accident is applied to explain this uncertain inverse traffic accident reconstruction, and the mean values of unknown impact parameters are assigned into the established simulation model to check the effectiveness of results. By matching the simulated and actual impact responses, it shows that the proposed approach is efficient and reliable for traffic accident reconstruction under uncertainty. This approach can help the accident investigators and judges reasonably evaluate the behavior of accident participants and complete the accident responsibility confirmation.