A Physics-Grounded Multi-Modal Sensor Fusion Framework for Pedestrian Impact Kinematic Reconstruction Under...

Area of Science:

• Engineering
• Forensic Science
• Signal Processing

Background:

• Pedestrian-vehicle collisions leave critical kinematic data irretrievable by the time of investigation.
• Accurate collision reconstruction is vital for forensic analysis and understanding injury mechanisms.

Purpose of the Study:

• To present a Phase 1 design for a multimodal sensor fusion and signal-processing framework for reconstructing pedestrian-vehicle collision kinematics.
• To evaluate the framework's performance using simulation and uncertainty analysis.

Main Methods:

• Utilized a multimodal sensor suite including 128-channel LiDAR, 1080p NIR stereo cameras, and a 2 kHz Inertial Measurement Unit (IMU).
• Implemented sensor fusion via Kalman filtering and signal processing using Savitzky-Golay polynomial differentiation.
• Employed Monte Carlo uncertainty propagation and sensitivity analysis on a simulated scenario.

Main Results:

• Under simulated conditions with ±0.5 m throw-distance uncertainty, velocity reconstruction yielded an uncertainty of ±2.03 km/h.
• Sensitivity analysis revealed that a 10% acceleration noise spike could amplify injury metrics by 26.9%.
• Demonstrated quantitative justification for noise-optimal pre-filtering in collision reconstruction.

Conclusions:

• The proposed bimodal kinematic-acoustic architecture offers a physically interpretable foundation for collision reconstruction.
• Phase 1 results provide a basis for further validation and development towards forensic application.
• A five-phase validation roadmap is proposed, emphasizing rigorous testing before forensic deployment.