DrivFluent - Making Mobility Fluent and Safe

Inspiration

Distracted driving is a growing and serious threat to road safety. It causes approximately 1.6 MILLION fatal accidents each year. According to the Canadian Council of Motor Transportation Administrators, a distracted driver is 23 TIMES more likely to be involved in a collision. Losing someone you love simply because a driver was inattentive is UNACCEPTABLE to us.

What it does

INTRODUCING DRIVFLUENT, a solution to tackle his problem to the very core. By leveraging Computer Vision technology, we monitor the behavior and attention of drivers while driving and providing real-time feedback on their behavior.

We can precisely locate where and when the driver was distracted and provide statistical visualizations.

We do so by placing an IOT device inside the car cabin which takes live video feed as input and geolocation of the user. The live video runs through our AI model and provides real-time feedback to the driver of their behavior. Our product is the first of its kind that can precisely detect 10 distracted driving behaviors during day AND NIGHT.

How I built it

We built a working prototype by training the core AI model developed using Deep learning frameworks Tensorflow and Keras on the state farm dataset which consists of 10 driving behavior classes. We wrapped the core model using Python-flask API which can take user video/images in real-time and provide driving behavior feedback to the user.

Challenges I ran into

The biggest challenge was to collaborate with team members remotely. But we could overcome this challenge by having recurrent meetings and sharing our progress with each other. The next big challenge was the infrastructure required to train such a huge deep learning model with 23000 images. We leveraged the google collab platform that provides free computation power and we could easily train our model there.

Accomplishments that I'm proud of

We are really proud of the deployment pipeline we have created and we created an API which user can directly use and get feedback from.

What I learned

I learned how deep learning models are deployed end-to-end for utilization by end-user.

What's next for DrivFluent - Making Mobility Fluent and Safe

Making our AI model more robust and mature. We plan on gathering more data for distracted behaviors exhibited by drivers while driving. And add it to our model to learn so it can address and tackle more types of behaviors.