TL;DR
This paper introduces a novel method for reconstructing car-running pedestrian accidents by employing a humanoid robot approach to simulate the pedestrian's dynamic running motion. This method improves the accuracy of accident reconstruction by accounting for the pedestrian's initial state, leading to more realistic simulations of collision kinematics and injury outcomes.
What Problem Does This Paper Address?
Car-pedestrian traffic accidents (TAs) are a significant cause of fatalities, and accurately reconstructing these events is crucial for accident investigation and prevention. A major challenge in TA reconstruction is accurately modeling the dynamic behavior of running pedestrians (RPs) at the moment of impact. Existing digital human body models (HBMs) either excel at simulating human movement (multibody HBMs) but lack injury prediction capabilities, or are highly precise for injury analysis (finite element HBMs) but cannot realistically simulate dynamic motion like running. This gap prevents a comprehensive understanding of how a pedestrian's running state influences collision dynamics and subsequent injuries.
Methods
The authors propose a "humanoid-robot-based reconstruction method" that combines the strengths of multibody (MB) and finite element (FE) HBMs.
- Humanoid Robot Theory for Pedestrian Motion: They adapt the principles of humanoid robot control to simulate the running motion of a digital pedestrian. This involves defining generalized coordinates for the virtual leg and deriving differential dynamic equations to capture the take-off and flight phases of running.
- Rigid-Flexible Coupled HBM: To achieve both accurate motion simulation and biomechanical injury analysis, they utilize a rigid-flexible coupled HBM. This model splits the human body into rigid multibody segments for efficient motion simulation and uses local FE models for critical body parts (like the lower limbs) to accurately predict injuries.
- MADYMO Software: The dynamic simulations, including the running process and the pedestrian-vehicle collision, were performed using MADYMO software, which supports rigid-flexible coupled dynamics.
- Accident Reconstruction Process: The reconstruction involves two stages:
- Pre-simulation of RP Motion: Using the humanoid robot method and accident information, the RP's dynamic running process is simulated to determine the joint states and velocities at the moment of collision.
- Pedestrian-Vehicle Collision Simulation: The initial conditions derived from the pre-simulation are then input into the pedestrian-vehicle collision model to reconstruct the accident.
Datasets
The paper uses a specific accident case for validation. The initial parameters for this case were derived from:
- Accident Video Footage: Captured images were analyzed using PhotoModeler software to determine the vehicle's velocity and the pedestrian's running speed.
- Forensic Autopsy Reports: The simulated injury results were compared against actual autopsy findings from the accident victim.
The paper also refers to existing experimental data and studies for validating the HBMs, including:
- TNO HBM: Used as a baseline for comparison.
- HUMOS HBM: Another standard FE HBM.
- Postmortem Human Subjects (PMHS) Tests: Used for validating head impact simulations.
- Cadaver Impact Tests: Used for validating limb and trunk impact responses.
Key Insights and Ideas
- The "Running State" Matters: The study strongly suggests that accurately modeling the pedestrian's initial running posture and velocity significantly impacts the accuracy of accident reconstruction, particularly in terms of injury prediction and the pedestrian's final resting position.
- Humanoid Robot Theory as a Bridge: Applying humanoid robot principles to digital human models provides a robust framework for simulating complex dynamic movements like running, which are difficult to capture with traditional methods.
- Rigid-Flexible Coupling is Key: The hybrid rigid-flexible HBM approach effectively balances the computational efficiency of multibody models with the biomechanical fidelity of finite element models, enabling both realistic motion and injury analysis.
- Injury Differences Highlighted: By comparing simulations with and without considering the initial running state, the authors demonstrate that neglecting this factor can lead to significant discrepancies in predicted injury severity and characteristics.
Key Results
- Accurate Reconstruction: The proposed method successfully reconstructed a car-running pedestrian accident case, with simulated dynamics closely matching video evidence and simulated injuries aligning well with forensic autopsy results.
- Injury Validation: The simulation accurately predicted severe head injuries (HIC = 3913, far exceeding the limit) and lower limb fractures (von Mises stress exceeding tolerance limits), consistent with autopsy findings.
- Impact of Initial State: Simulations comparing different initial states (running vs. static) showed notable differences in the pedestrian's rest position and final impact location, with the running state leading to a more accurate representation of the accident.
- Relative Error Analysis: Case 1 (considering running dynamics) showed lower relative errors in HIC and lower limb stress compared to cases that ignored the initial motion state, highlighting the improvement in accuracy. For instance, the relative error in HIC between Case 2 (static posture, initial velocity) and Case 1 was 9.4%, and for lower limb stress, it was 47.0%.
Limitations and Future Work
The authors acknowledge several limitations:
- Two-Stage Reconstruction: The current process requires separate simulation stages for RP motion and collision, which can be cumbersome.
- Limb Trajectory Accuracy: While velocities can be accurately applied, ensuring the precise trajectories of arms and limbs in the HBM match real-world movements across different speeds remains challenging.
- Biological Variability: The HBM, though scaled, cannot fully account for individual biological differences in biomechanics.
- Pre-Collision Reactions: The study did not consider pre-collision reactions like driver braking or pedestrian evasive maneuvers, which can influence accident outcomes.
Future work could focus on integrating these aspects into a more unified simulation framework, improving limb trajectory modeling, and incorporating more sophisticated pre-collision behaviors.
Summary updated