YOLOv8n + CNN + Grad-CAM + Fuzzy Logic for Drowsiness Detection

A two stage pipeline for driver drowsiness detection. YOLOv8n detects the face region. A CNN predicts active vs drowsy. Grad-CAM shows which region influenced the CNN output. Fuzzy logic converts the probability into risk levels to reduce false alarms.

---

Folder Structure

|-- results/
|   |-- gradcam_samples/
|   |-- report.txt
|   |-- confusion_matrix.png
|-- src/
|   |-- Drowsiness Detection.ipynb
|-- docs/
|   |-- methodology.md
|   |-- train.md
|   |-- dataset.md
|   |-- inference.md
|-- requirements.txt
|-- README.md

Quick Links

• Dataset details: docs/dataset.md
• Full methodology: docs/methodology.md
• Training steps: docs/train.md
• Inference steps: docs/inference.md

Results (from this repo)

Confusion Matrix

Confusion Matrix After Fuzzy Logic

Grad-CAM Samples

Sample	Preview
1
2

Metrics Text

See: results/report.txt

Notes

• All steps are implemented in src/Drowsiness Detection.ipynb.
• The docs explain the same pipeline in paper friendly form.