Related Experiment Video
Updated: Jan 8, 2026
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
11.1K
Pedestrian Injury Case Reconstruction through Data Fusion and Machine Learning
Xiaoyang Song1, Wenbo Sun2, Jingwen Hu2
1Department of Industrial and Operations Engineering, University of Michigan.
Stapp Car Crash Journal
|December 22, 2025
Summary
Machine learning models can now reconstruct missing vehicle speed data in pedestrian crashes. This improves pedestrian injury analysis and safety research by combining simulation and real-world data.
Area of Science:
- Road Safety
- Biomechanical Engineering
- Data Science
Background:
- Pedestrian injuries from motor vehicle crashes are rising in the U.S.
- Existing police data lacks complete crash and injury details, hindering research.
- Accurate data is crucial for developing effective pedestrian protection strategies.
Purpose of the Study:
- To develop a machine learning approach for imputing missing crash data in pedestrian incidents.
- To combine simulation and field data for enhanced pedestrian crash analysis.
- To improve the accuracy of pedestrian injury risk modeling.
Main Methods:
- Generated 9,000 MADYMO simulations with varying crash parameters.
- Trained Gaussian process (GP) surrogate models to predict injury risks.
- Used maximum likelihood estimation to impute missing vehicle speeds in trauma data.
Main Results:
- Imputed vehicle speed distribution closely matched the Pedestrian Crash Data Study (PCDS) dataset.
- Reconstructed vehicle speeds in CIREN cases showed a small average deviation (9 kph) from physics-based methods.
- Predicted injury risks aligned with observed Abbreviated Injury Scale (AIS) levels.
Conclusions:
- Machine learning effectively reconstructs missing crash data, particularly vehicle speed.
- The proposed method enhances pedestrian injury risk modeling and analysis.
- This approach supports advancements in pedestrian protection research.
