Research Highlights

Vulnerability Analysis: Identifies pedestrians in non-upright postures as a critical blind spot in current ADAS, representing ~7.8% of Japanese traffic fatalities.
Architectural Innovation: Proposes the Advanced Falling Object Detection System (AFODS) using YOLOv7-Tiny and MFCC-based acoustic verification.
Regulatory Alignment: Maps detection requirements for fallen road users to ISO 26262 Automotive Safety Integrity Levels (ASIL).
Ethical Engineering: Replaces utilitarian “Trolley Problem” models with a deontological framework prioritizing vulnerable road users.
IP Disclosure: System architecture protected under Japanese patent application 2025-167440.

Abstract

Modern automotive safety relies heavily on upright-pedestrian detection, leaving individuals in compromised postures—due to medical emergencies or prior impacts—largely “invisible” to standard sensors. This paper argues for an integrated safety paradigm that synthesises ISO 26262 functional safety protocols with multi-modal AI. By deploying the Advanced Falling Object Detection System (AFODS), which utilises spatial recognition and acoustic signal processing, we demonstrate a significant increase in detection reliability for prone individuals. This technological shift is supported by a deontological ethical framework that mandates the protection of the most vulnerable participants in the transit ecosystem.

The Perception Gap in Modern Transit

Despite the ubiquity of Advanced Driver-Assistance Systems (ADAS), a dangerous “classification gap” persists in current vehicle perception logic. While systems excel at identifying standard pedestrians, humans in non-upright postures are often misinterpreted as road irregularities. Statistical evidence from Japan indicates that pedestrians account for roughly 7.8% of traffic fatalities [1]. In low-light environments, the True Positive Rate (TPR) for these individuals can drop to 21.4%, highlighting a critical need for sensory evolution [2].

Functional Safety and System Integrity (ISO 26262)

To move beyond reactive safety, vehicle perception must be treated as a high-integrity functional requirement under the ISO 26262 standard.

a) ASIL Classification: We argue that “fallen pedestrian” detection should be a high-severity focal point in Hazard Analysis and Risk Assessment (HARA) to determine appropriate Automotive Safety Integrity Levels (ASIL).
b) Sensory Envelope: Achieving fail-operational behavior requires a fusion of Long-Wave Infrared (LWIR), Near-Infrared (NIR), and Ultrasonic sensors.
c) Thermal Verification: LWIR is indispensable, as it detects the consistent biological thermal signature (36.5–37.5°C) regardless of lighting or body orientation [2].

The AFODS Technical Pipeline

The Advanced Falling Object Detection System (AFODS) introduces a redundant AI architecture to bridge the perception gap:
a) Spatial Identification: A YOLOv7-Tiny model optimised for edge devices provides real-time detection of low-profile human signatures [2].
b) Predictive Kinematics: By utilising Recurrent Neural Networks (RNNs), the system can interpret staggering motions to forecast a fall before it occurs.
c) Acoustic Confirmation: To mitigate false positives, the system employs Mel-Frequency Cepstral Coefficients (MFCC) to verify the distinct acoustic signature of a physical impact or fall [3].

Deontological Ethics in Autonomous Logic

Automotive AI development often centres on the “Trolley Problem,” yet this utilitarian approach is frequently criticised as an unrealistic edge case.
a) Duty of Care: We advocate for deontological ethics, which posits an inherent duty to protect those with limited mobility.
b) Statistical Prevention: By achieving a 98.2% detection rate, AFODS shifts the focus from choosing between harms to proactive avoidance [2,4,5].
c) Transparency and Accountability: Multi-modal verification provides an explainable audit trail for breaking decisions [6].

Conclusion

Advancing road safety requires a transition from forensic observation to preventative engineering. By aligning sensor training with biomechanical data and rigorous functional safety standards, the “invisible” road user can be brought into the light of machine perception [7].

Intellectual Property & Disclosure

Patent Information: The systems and methodologies described herein, including the AFODS architecture, have been filed under Japanese patent application number: 2025-167440, Filing date: 3 October 2025.

Conflict of Interest Statement: The author is the Chairman & CEO of AN Holdings Co., the entity managing the development of the technology discussed. This research was conducted in coordination with the Shiga University of Medical Science.

References

1. Hitosugi M, Tokudome S, Yokoyama T (2021) Factors influencing fatalities or severe injuries to pedestrians lying on the road in Japan: A nationwide police database study. Healthcare 9(11): 1433.
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): 149-165.
3. Rezaul KM, Jewel M, Islam MS, Barua N, Rahman MA, et al. (2024) Enhancing audio classification through MFCC feature extraction and data augmentation with CNN and RNN models. International Journal of Advanced Computer Science and Applications 15(7): 44-55.
4. Bonnefon JF, Shariff A, Rahwan I (2016) The social dilemma of autonomous vehicles. Science 352(6293): 1573-1576.
5. Goodall NJ (2021) Away from trolley problems and toward risk management. Ethics and Information Technology 23(3): 435-445.
6. European Commission (2020) Ethics of connected and automated vehicles: Recommendations on road safety, privacy, fairness, explainability and responsibility. Publications Office of the European Union.
7. Wang M (2025) The evolution of autonomous driving technology and its ethical challenges: A pedestrian-first perspective. Proceedings of the 2^nd International Conference on Data Science and Engineering pp. 88-94.