As new drivers, we were surprised to learn the following: 1) Distracted driving is responsible for 58% of teen crashes 2) Around 400k injuries each year in the US alone are caused because of distracted driving 3) It takes 3 seconds of distraction to cause a fatal crash We know that distracted driving is completely preventable. We also found out that there were no apps on the market that currently analyze images in-depth and report the different types of distraction one could experience when driving.

What it does is an easy-to-use iOS application that implements custom image classification algorithms to detect specific distracted behavior in drivers and alert them in real-time with an auditory warning. Concretely, can detect 10 different states of the driver, including safe driving, texting while driving, and drinking while driving. After completing their drive, the driver is presented with relevant statistics, such as their primary distraction states, to improve in the future. Furthermore, to ensure its accessibility, we have configured's image classification algorithms to run locally and thus it does not require internet access.

How we built it

We used Machine Learning, specifically transfer learning built on top of VGG 16, to build a neural network that could detect over 10 different types of distractions. We then meshed our machine learning algorithms with

Challenges we ran into

Building the initial UI was challenging as we had to strive for an intuitive feel. However, the hour that we spent on this was well worth it as it increased the clarity of our initial vision for

Accomplishments that we're proud of

93%+ validation accuracy of our machine learning model, being able to combine the machine learning and mobile app frontend.

What we learned

How to serve a machine learning model for computing on the edge!

What's next for

We will continue to develop and finetune the app!

Built With

Share this project: