SafeDrive.AI

Inspiration

The recent statistics of millions of North Americans who die unnecessarily every year because of impaired driving. Over 75% of all fatal car crashes involve at least one impaired driver. By focusing on the steps before an impaired driver gets into their car, we would be able to drastically reduce this number.

What it does

This model is based on neuroplasticity games to identify changes in reaction time and quality. First there is a facial recognition test that is based on a convolutional neural network that we built from scratch. After, there is a speech test that used another convolutional neural network we created to identify slurs in speeches as well as changes in pitch and frequency using natural language processing. Finally, there is a trio of neuroplasticity games to test for reaction times and skill testing math questions to ensure that the driver is not impaired. It should be mentioned that all results are compared to those that the driver has achieved when they were not impaired, as a baseline for comparison.

How I built it

We built the machine learning model to identify if the user's face was impaired first because that required the most work. We had to train the model with thousands of datasets we found on kaggle. After training and running the eye test, we moved on to the speech test. This was made using natural language processing to identify the changes in pitch and frequency of the users speech from a pre-recorded baseline. Finally, we used react to build the games and skill testing questions.

Challenges I ran into

Training the neural networks and finding large datasets with valuable pictures and audio files proved to be a problem. However, with the help of Google, it was accomplished.

Accomplishments that I'm proud of

For some team members, it was their first time working with machine learning at all. For all team members, it was our first time creating a machine learning mode for audio files.