From Detection to Forensics: An Integrated Safety Architecture for Fallen Pedestrian Protection

Fallen pedestrians—those lying prone, supine, or crouching on roadways—represent a critical and under-addressed vulnerability in contemporary Advanced Driver Assistance Systems (ADAS). In simulation, a baseline RGB camera system achieved a 21.4% true-positive detection rate (TPR) at night for non-upright subjects, compared with a simulated 98.2% TPR achieved by the Advanced Fallen Object Detection System (AFODS) under the same night-time (0 lux) condition—a 76.8 percentage-point simulated detection gap. This article synthesises three complementary peer-reviewed studies into a proposed closed-loop safety architecture: (1) simulation-based multi-modal detection via AFODS; (2) a physics-grounded design evaluated through simulation for post-collision kinematic reconstruction; and (3) simulation-based injury-risk quantification translating detection latency into Head Injury Criterion (HIC) and AIS-graded fatality probability. The integrated framework is associated with Japanese Patent Application No. 2025-167440, filed on 3 October 2025, and JST support has been approved for the planned PCT filing and further development of the detection and information-processing claims. In the source simulation, modelled fatal head-injury probability at 50 km/h falls from 66.2% (no-detection baseline) to 0.7% under worst-case AFODS-triggered braking—a 65.5-percentage-point model-chain contrast that has not yet been empirically observed. A five-stage validation roadmap is proposed, targeting eventual conformance assessment against ISO 26262, ISO 21448 (SOTIF), and relevant NCAP protocols. The article identifies open research challenges and a proposed trajectory toward prototype testing; it does not report empirical validation, legal admissibility findings, or deployment, all of which remain future work.