|
Abstract
|
OBJECTIVE: The aim of this study is to identify the effects of pedestrian physique differences on head injury prediction in car-to-pedestrian accidents via deep learning.
METHODS: A series of parametric studies was carried out using a family car finite element model and MADYMO pedestrian models (AM50, AF05, 6YO). The car model was developed and tuned by 11 impact tests. The initial gaits for the pedestrian models were obtained from volunteer experiments to reproduce 420 pre-crash reactions. Furthermore, by factoring the pedestrian models (3 types), pedestrian directions (2 each), impact positions (3 each), and car velocities (6 levels) with the pre-crash parameters, a total of 45,360 car-to-pedestrian impact simulations were performed. After the simulations, image datasets were created by labeling the pedestrian collision images with head injury criteria of 15 ms (HIC) and dividing the images into training and test data based on model type. Next, deep learning was conducted using the training dataset to obtain trained models. Finally, the effects of pedestrian physique differences on head injury predictions were investigated based on the accuracy of each trained model for test data.
RESULTS: The results indicate that the head impact area and the amount of pedestrian information in the image differ depending on the pedestrian models. In head injury prediction with deep learning, AF05 showed the highest prediction accuracy (93.25%), followed by AM50 (90.61%) and 6YO (88.29%). These results using deep learning show that pedestrian physique differences affect the head injury prediction accuracies by 2.32-4.96 points.
CONCLUSIONS: Based on the prediction results of the trained models that learned the relationships between the pedestrian collision images and HIC from simulations, we demonstrated the desirable performance of deep learning methods in head injury prediction for adult men, women with small physique, and children. Furthermore, our results confirmed the effect of pedestrian physique differences on the injury prediction accuracy.
Language: en |