Sleepy Savior

Model detecting distraction by phone

Inspiration

As aspiring teenagers with driver licenses and permits, we wondered more about the constant dangerous side-effects of falling asleep on the wheel. Combining our curiousity and our willingness to help others, we created Sleepy Savior, to save drivers from their worst nightmare.

What it does

Our machine learned computer model uses a webcam to detect factors of drowsy drivers, such as their head tilted down to their movement of their eyes, to alarm them to be alert and ready on the wheel in case of an accident.

How we built it

We utilized the AMD developer program's fast and efficient gpu that allowed us to work on complex. and large data sets especially of the different positions of drowsy-related symptoms. The model picks up on tiny-land marks around the eyes, as well as the mouth and general position of the head to make accurate estimates of whether a driver is concious or not.

Challenges we ran into

We had difficulty being accurate with augments such as hats and other facial apparel because face pixels are covered or disfigured creating more uncertainty with the model

Accomplishments that we're proud of

Created an ML facial recognizer that can accurately determine and wake up drowsey faces and micro sleeps with 89% accuracy. We also leveraged higher end gpus to create extremely detail oriented model that can render facial drowsiness despite augments such as blur, darkness and sunglasses

What we learned

Learned how to cooperate on github and utilize databases and cloud service gpus to leverage higher level machine learning

What's next for Sleepy Savior

Next steps would be to integrate this into dashcams which are usually looking at a drivers face and eventually add incorporate other active features such as a moving steering wheel.