Abstract: Considering that ground related injury is difficult to predict, a prediction method for pedestrian safety/danger landing mechanisms in pedestrian-vehicle collision is proposed. Firstly, 140 reconstructed real pedestrian-vehicle collision cases are selected and relevant data before, during and after the collision are extracted. 140 sets of data containing 12 parameters are obtained by means of significance and collinearity tests. Then, 10 sets of models containing 3 prediction models before, during and after the collision are established by stepwise regression method, and 1 set of optimal models is selected as the prediction model of pedestrian safety/danger landing mechanism. The results show that all 10 sets of models can predict pedestrian safety/danger landing mechanisms well, indicating that pedestrian landing mechanisms are predictable; The average prediction accuracy of the three models in the selected optimal model set are 76.5%, 82.7%, and 87.8% with the required prediction parameters of 1, 3, and 4, respectively. Further analysis reveals that the three models in the optimal model set can also be well applied in MADYMO simulation, and the Ratio of Pedestrian height to Bonnet leading edge height(R_P-B) will affect the model prediction performance. Further, an improved model considering the R_P-Bis proposed, and it is found that the improved model can significantly improve the average prediction accuracy of the in-collision model.

Key words: pedestrian-vehicle collision, ground related injury, prediction of pedestrian landing mechanism, logistic regression