From Post-Mortem to Prevention: Redefining “Invisible” Pedestrians Through ISO 26262 and Multi-Modal AI

Introduction: The Forensic Imperative

In forensic pathology, the reconstruction of pedestrian-vehicle collisions often reveals a critical disparity: while modern vehicles are adept at avoiding “standard” road users, they remain remarkably blind to individuals in compromised postures. Nationwide database studies in Japan indicate that pedestrians lying on the road account for approximately 7.8% of all traffic fatalities [1].

From a biomechanical perspective, a fallen human presents a radically different cross-section that monocular camera systems often misinterpret as road irregularities. This “classification gap” is a silent contributor to secondary impact fatalities, particularly in low-light conditions, where standard systems yield a 21.4% True Positive Rate (TPR) [2]. The forensic community must therefore advocate for a transition from observing the aftermath to engineering active prevention.

Functional Safety and ISO 26262: A Forensic Requirement

The detection of fallen pedestrians is no longer just a luxury feature; it is a critical safety goal under the ISO 26262 international standard for functional safety.

• ASIL and Risk Assessment: ISO 26262 mandates a rigorous approach to Automotive Safety Integrity Levels (ASIL). Our research suggests that the “fallen pedestrian” should be considered a high-severity hazard in Hazard Analysis and Risk Assessment (HARA).

• Multi-Modal Redundancy: To comply with ISO 26262’s demand for fail-operational behaviour, we advocate for a “sensory envelope” combining Long-Wave Infrared (LWIR), Near-Infrared (NIR), and Ultrasonic sensors. LWIR is essential because it captures the biological thermal signature (36.5–37.5°C), which remains constant regardless of posture or lighting [2].

Deep Technical Integration: YOLOv7-Tiny and RNNs

Our Advanced Falling Object Detection System (AFODS) addresses the detection gap through a custom AI pipeline designed to handle the unique visual signatures identified in forensic case studies:

• Detection: Utilising a YOLOv7-Tiny architecture optimised for edge computing, the system identifies prone individuals with high precision [2].

• Prediction and Acoustic Verification: By integrating Recurrent Neural Networks (RNNs), the system analyses the kinematics of a stagger to predict a fall before it occurs. This is supplemented by audio classification using Mel-Frequency Cepstral Coefficients (MFCC) to identify the acoustic signature of a fall, providing a secondary layer of confirmation [3].

Ethical Implications: AFODS and the "Trolley Problem"

The “Trolley Problem”—a moral dilemma where an AI must choose between two harms—is often criticised as an unrealistic “edge case.” However, in the context of AFODS, it becomes a tangible engineering challenge [4,5].

• Prioritising the Vulnerable: Our research moves beyond utilitarianism toward deontological ethics, which emphasises the inherent duty to protect the most vulnerable road users. A fallen pedestrian is inherently more vulnerable as they lack the mobility to self-evacuate.

• Deterministic vs. Probabilistic Ethics: AFODS shifts the focus from “who to hit” to “how to avoid.” By achieving a 98.2% detection rate, the system reduces the statistical likelihood of entering a “Trolley” scenario in the first place [2].

• Transparency: As mandated by the EU Expert Group on Ethics of Connected and Automated Vehicles, safety systems must be explainable. AFODS’s use of multi-modal confirmation provides a clear “forensic audit trail” for why a braking decision was made [6].

Conclusion

The future of forensic science is not merely the study of death, but the engineering of survival. By demanding that automotive manufacturers align their AI training with forensic biomechanical data and ISO 26262 standards, we can eliminate the “invisible” status of fallen road users. The transition from forensic observation to proactive, AI-driven intervention is the most significant leap traffic medicine has taken in decades [7].

References

1. Koh, M., Hitosugi, M., Kagesawa, E., Narikawa, T., & Takashima, K. (2021, October). Factors influencing fatalities or severe injuries to pedestrians lying on the road in Japan: Nationwide Police Database Study. In Healthcare (Vol. 9, No. 11, p. 1433). MDPI.
2. Barua, N., & Hitosugi, M. (2025). Advanced Multi-Modal Sensor Fusion System for Detecting Falling Humans: Quantitative Evaluation for Enhanced Vehicle Safety. Vehicles, 7(4), 149.
3. Rezaul, K. M., Jewel, M., Islam, M. S., Barua, N., Rahman,M. A., Bin Sulaiman, R., ... & Tuj Asha, U. F. (2024). Enhancing Audio Classification Through MFCC Feature Extraction and Data Augmentation with CNN and RNN Models. International Journal of Advanced Computer Science & Applications, 15(7).
4. Bonnefon, J. F., Shariff, A., & Rahwan, I. (2016). The social dilemma of autonomous vehicles. Science, 352(6293), 1573-1576.
5. Goodall, N. J. (2016). Away from trolley problems and toward risk management. Applied Artificial Intelligence, 30(8), 810-821.
6. Planet, C., & Aguinaga, C. J. F. (2020). Ethics of Connected and Automated Vehicles Recommendations on road safety, privacy, fairness, explainability and responsibility.
7. Wang, M. (2025). The evolution of autonomous driving technology and its ethical challenges: A pedestrian-first perspective. Proceedings of the 2nd International Conference on Data Science and Engineering, 88–94.