A real-time AI-powered system to detect driver or user drowsiness using facial and behavioral cues such as prolonged eye closure, yawning, and head pose dynamics.
The project leverages computer vision and deep learning to enhance road safety, workplace productivity, and health monitoring.
this is a collaborative team project
- Real-Time Detection of drowsiness using webcam input.
- YOLOv5 for facial detection and landmark localization.
- Eye Aspect Ratio (EAR) & Mouth Aspect Ratio (MAR) for blink/yawn detection.
- LSTM networks for temporal sequence learning (reduces false positives).
- Custom alert system with visual, audio, and haptic feedback.
- Modular design, adaptable for desktop, embedded devices (Raspberry Pi, Jetson Nano), and cloud deployment.
- Programming Language: Python
- Libraries & Frameworks: PyTorch, OpenCV, MediaPipe/dlib, Matplotlib
- Models: YOLOv5, CNN, LSTM
- Deployment: Local (desktop/embedded), optional cloud support
- Custom dataset (recorded with webcam under varied conditions)
- Capture live video input from webcam.
- Detect face and landmarks using YOLOv5 + MediaPipe/dlib.
- Compute EAR, MAR, and head pose angles.
- Feed sequential features into LSTM network for drowsiness classification.
- Trigger alerts (visual/audio/haptic) when drowsiness is detected.
- Robust detection across varied lighting and user conditions.
- Temporal analysis with LSTM reduced false positives compared to single-frame methods.
- Potential applications in road safety, healthcare, and workplace monitoring.
- Incorporation of transformer-based models (Vision Transformers).
- Multimodal data fusion with physiological signals (heart rate, motion).
- Cloud-based analytics dashboard with federated learning.
- Aryan Shrivastav
- Kshitij Pratap Tomer
- Aaryak Bhargava
- Aryaman Jain
This project is developed for academic purposes. For commercial use, please contact the contributors.