Initialize Stream
Architecture
Built a dual-stream processing pipeline where interior (driver) and exterior (road) camera feeds are ingested via OpenCV, processed by deep learning models, and fused in a shared event bus. Results are transmitted to a mobile frontend (Flutter) using high-frequency WebSocket messaging.
Implementation
Architecture involves integrating YOLOv8 and MediaPipe for simultaneous processing, optimizing model weights for mobile-compatible inference speeds, and building a low-latency state synchronization layer between the backend and hybrid app.
Outcome
Provides a functional prototype for safety-critical driver assistance, focusing on tight model-to-product integration and multi-sensor data fusion.
What it does
- Real-time Fatigue, Drowsiness, and Distraction monitoring via MediaPipe
- Object detection for road hazards (Stop signs, Pedestrians, Potholes) using YOLOv8
- Dynamic risk scoring algorithm based on fused behavioral telemetry
- Automated emergency protocol routing for critical safety incidents
- Optimized inference loop for low-latency (<300ms) mobile performance