WiseMotion
Best Use of Computer Vision at Hack the North. Real-time driver-safety monitor.
The problem
Drowsiness and poor posture both make drivers slower to react, and neither is easy to notice in yourself. The goal was a monitor that runs on a normal webcam and warns the driver out loud, fast enough to matter.
Constraints
- Built in a hackathon weekend.
- Must run in real time on a laptop webcam, with no special hardware.
- Alerts must come quickly, and normal blinking must not trigger them.
How it works
Webcam frames: OpenCV captures frames from a standard webcam.
Key decisions
Geometry before deep learning
A landmark-based Eye Aspect Ratio is fast, explainable and easy to tune on the spot, which mattered more under hackathon time limits than a heavier classifier.
Audio, not visuals
A driver should not have to look at a screen, so every alert is spoken.
Results
- 92% drowsiness precision
- Under 200 ms from detection to spoken alert
- Won Best Use of Computer Vision at Hack the North
What I'd do next
Calibrating the threshold per driver in the first minute, since resting eye shape varies from person to person.