A high-speed Computer Vision system designed for edge deployment, providing immediate auditory alerts based on real-time periorbital monitoring and facial landmark analysis.
- Facial Landmark Localization: Utilizes DLib’s 68-point predictor for high-precision eye and mouth tracking.
- EAR (Eye Aspect Ratio): Implemented mathematical thresholds to detect microsleep events based on eyelid closure duration.
- OpenCV Integration: Optimized frame-by-frame processing to maintain a low memory footprint.
- Inference Latency: Optimized the pipeline to achieve a response time of <30ms per frame, critical for life-saving safety alerts.
- Accuracy: Achieved 95%+ accuracy in varied lighting conditions and head poses.
- Embedded Readiness: Architected specifically for low-power edge devices (e.g., Raspberry Pi or Jetson Nano).