A prediction-based framework for developing integrated pedestrian safety systems under high-risk scenarios.

Area of Science:

• Automotive Safety Engineering
• Intelligent Transportation Systems
• Biomechanical Engineering

Background:

• Pedestrians are highly vulnerable road users, necessitating advanced safety measures in intelligent vehicles.
• Current active hood systems rely on contact sensors, limiting deployment time in critical scenarios.
• Mitigating pedestrian injuries in unavoidable collisions requires improved pre-impact response.

Purpose of the Study:

• To develop and validate a prediction-based integrated pedestrian safety framework for enhanced active hood deployment.
• To improve pedestrian safety by shifting from reactive impact detection to proactive pre-impact prediction.
• To provide a sufficient time margin for active hood deployment before pedestrian head contact.

Main Methods:

• Constructed a vehicle-perspective, human-in-the-loop virtual reality dataset with high-risk pedestrian scenarios.
• Developed a short-horizon forecasting model using causal CNN and stacked LSTM for pedestrian behavior prediction.
• Evaluated the framework using simulations and real-world near-miss and collision cases.

Main Results:

• The forecasting model achieved sub-meter accuracy in predicting abrupt pedestrian movements.
• The prediction-based system offered a 200 ms forecast horizon, enabling full active hood deployment before head contact.
• Simulations showed up to 62.1% reduction in Head Injury Criteria (HIC) at 45 km/h.
• Real-world evaluations confirmed trigger conservatism with zero false positives in near-miss cases.

Conclusions:

• The proposed framework enables a paradigm shift towards proactive pedestrian protection systems.
• Early prediction and deployment of active hoods significantly reduce pedestrian head injury severity.
• The system demonstrates robustness, transferability, and potential for reversible electric actuation.