SleepyWheels

An ensemble drowsiness detection system

This project is an effort towards higher accuracy and more robustness to various real-life situations (different camera angles, skin complexion, presence of face wear like eyeglasses/masks, etc.) in which a driver's drowsy behaviour must be detected and they must be alerted before a road accident can occur. It consists of:

• an EfficientNetV2B0 CNN model that outputs the probability of the driver being sleepy
• a FaceMesh landmark detector (from Google's Mediapipe library) that enables calculation of aspect ratios of eyes and mouth
• an alarm that gets triggered if the driver is sleepy
• a web dashboard that can be visited later to check alarm and yawn timestamps from previous journeys

To run the system locally

1) Install python version 3.9 2) Create a new python virtual environment and activate it 3) pip install -r requirements.txt 4) python ./app/app.py