AFODS Operational Sequence Documentation

Authors: Dr. Nick Barua · Prof. Masahito Hitosugi
Department of Legal Medicine, Shiga University of Medical Science, Otsu, Shiga, Japan
Supporting: MDPI Vehicles 2025 · DOI: 10.3390/vehicles7040149
Patent Filed: Japanese Patent Application No. 2025-167440 (Filed: 3 October 2025)

---

📌 Overview

This repository provides supplementary documentation and the core diagram for the operational sequence of the Advanced Falling Object Detection System (AFODS), supporting the peer-reviewed manuscript:

Advanced Multi-Modal Sensor Fusion System for Detecting Falling Humans: Quantitative Evaluation for Enhanced Vehicle Safety
Vehicles, MDPI, 2025, 7(4), 149 · DOI: 10.3390/vehicles7040149

The Graphical Abstract and the five-stage processing diagram illustrate the advanced data processing pipeline designed for proactive threat detection — detailing the sequential flow from initial sensor data acquisition to the final decision and action stage. Validated across 320 controlled trials, AFODS achieved 98.2% TPR at night (0 lux), where the baseline visible-spectrum system collapsed to 21.4%.

---

🖼️ Five-Stage AFODS Pipeline

Stage	Name	Key Operations
1	Acquisition & Preprocessing	LWIR Thermal · NIR Stereo · Ultrasonic · Confidence-weighted fusion dynamically adjusts sensor trust based on environmental conditions
2	Detection & Classification	YOLOv7-Tiny (trained on 15,000+ images) · Hypothermia Detection Mode adjusts thermal threshold dynamically by ambient temperature
3	Depth Estimation & Motion Analysis	SGM stereo algorithm → dense disparity map · Lucas–Kanade optical flow → vertical motion tracking · Lightweight pose estimation distinguishes collapse from non-critical actions
4	Predictive Threat Assessment	GRU-based RNN analyses pose sequences over 1–2 s window · Detects pre-fall indicators (staggering, loss of balance) before collapse is complete
5	Decision & Action	TTC = D / v_vehicle · Braking triggered when TTC < 1.2 s AND confidence > 95% · Acoustic sensor as final corroborating factor · Target: ISO 26262 ASIL B · Mean latency: 46.3 ms (SD = 4.1 ms)

---

🔗 Repository Resources

Resource	DOI / Link	Purpose
Primary Codebase	Advanced-Multi-Modal-Sensor-Fusion-System-for-Detecting-Falling-Humans	Full system documentation, source code, and figures
Sensor Fusion Code	AFODS-Sensor-Fusion-Code	YOLOv7-Tiny and GRU model scripts
Video Demonstration	10.5281/zenodo.17460755	Real-time system performance demo
Methodology Figures	10.5281/zenodo.17621800	Key diagrams and results charts
Published Paper	10.3390/vehicles7040149	Full peer-reviewed methodology

---

📊 Validated Performance (320 Controlled Trials)

All values are from the peer-reviewed publication (Tables 1 & 2). Each condition repeated 20 times using standardised ATDs at 20 m.

Environmental Condition	AFODS TPR (%)	Baseline TPR (%)	p-value
Clear Daylight	99.5	96.8	0.041
Night (0 lux)	98.2	21.4	<0.001
Rain (50 mm/h)	96.4	55.7	<0.001
Fog (<50 m visibility)	95.8	32.1	<0.001

Mean System Latency: 46.3 ms (SD = 4.1 ms)
Mean Detection Range: 41.5 m (SD = 4.8 m) vs. Baseline 22.3 m (SD = 12.5 m)
False Positive Rate: AFODS avg. 1.6/24 h vs. Baseline avg. 32.2/24 h (95.0% reduction)

---

💡 How to Use

The flowchart and figures are intended for use in:

• Academic presentations and conference posters
• Technical documentation and reports
• Publications referencing the AFODS architecture
• ISO 26262 ASIL B safety case documentation

---

🔗 Related Publications

This repository is part of a unified research program on sensor fusion and road safety:

#	Title	Venue	Role
1	Advanced Multi-Modal Sensor Fusion System for Detecting Falling Humans	MDPI Vehicles	Technical foundation & benchmarks
2	Integrated Safety Architectures: Leveraging Multi-Modal AI and ISO 26262 to Protect Vulnerable Road Users	SSRN	System-level ISO 26262 safety architecture
3	From Post-Mortem to Prevention: Redefining "Invisible" Pedestrians through ISO 26262 and Multi-Modal AI	SSRN	Problem framing & ISO 26262 compliance
4	Sudden Incapacitation or Death at the Wheel: Probabilistic Risk Factors for Catastrophic Multi-Vehicle Collisions	SSRN	Epidemiological evidence for ADAS mandate
5	The Invisible Victims of the Road: Why ADAS Cannot See the Pedestrians Most Likely to Die	Preprints.org (under review)	AFODS forensic injury translation & regulatory advocacy
6	A Physics-Grounded Multi-Modal Sensor Fusion Framework for Pedestrian Impact Kinematic Reconstruction Under Uncertainty: Phase 1 Design and Theoretical Evaluation	MDPI Sensors	Forensic reconstruction companion — retrospective kinematic reconstruction from post-impact scene observables

---

📂 Related Repositories

• Advanced-Multi-Modal-Sensor-Fusion-System-for-Detecting-Falling-Humans — Primary repository with full system documentation and figures
• AFODS-Sensor-Fusion-Code — YOLOv7 and GRU model implementation scripts
• From-Post-Mortem-to-Prevention-AFODS — ISO 26262-aligned conceptual framework

---

📝 Citation

@article{vehicles7040149,
  author    = {Barua, Nick and Hitosugi, Masahito},
  title     = {Advanced Multi-Modal Sensor Fusion System for Detecting Falling Humans:
               Quantitative Evaluation for Enhanced Vehicle Safety},
  journal   = {Vehicles},
  volume    = {7},
  number    = {4},
  pages     = {149},
  year      = {2025},
  doi       = {10.3390/vehicles7040149},
  url       = {https://doi.org/10.3390/vehicles7040149}
}

---

📜 License

This project is licensed under the Apache 2.0 License — see the LICENSE file for details.

The system described in this repository is subject to Japanese Patent Application No. 2025-167440 (Filed: 3 October 2025).