RoadAI

• 

  RoadAI Logo

• 

  Launch Screen

• 

  Instructions

• 

  Detection Settings

• 

  Detection Screen

• 

  RoadAI Ad

Inspiration

Over the next 10 years, more and more self-driving cars will hit our roads. It is going to be a revolution in transportation. But according to recent studies, only 21% of people fully trust self-driving cars. With RoadAI Android app we decided to fill that gap. Using Artificial Intelligence, RoadAI intends to make people trust their transportation to self-driving cars by showing them how their phone and consequently a self-driving car perceives the world around it, acting as an intelligent co-driver. Simple setup a phone holder for cars and your phone. Self-driving cars promise safer roads for everyone and we decided to demonstrate that as well. In every driver’s life, there have been incidents where they have missed a red traffic light or stop sign. RoadAI detects and vocally warns the driver of potential dangers making the driving experience much safer.

What it does

Your phone detects red traffic lights, stop signs, cars and pedestrians, warns you and makes your driving experience safer. Users can also share their ride footage and contribute in making the model even smarter and more robust.

How I built it

Using PyTorch framework from Facebook and public datasets, we trained an AI model to detect Red Traffic lights, cars, pedestrians and stop signs. Then we converted the model to fit into an android device and we built the app around it. Let me show how it works.

Challenges I ran into

To compress the model for Android deployment

Accomplishments that I'm proud of

I am proud for both the App and the PyTorch model. I think this app will contribute into making people trust self-driving cars and artificial intelligence.

What I learned

To use PyTorch and deploy it into an Android app

What's next for RoadAI

Further steps would include a better and more robust model by collecting more data and using a more sophisticated architecture. To classify between all four classes in the future. Finally, when we reach the desired level of accuracy we will release of the app on google play.

Built With

• android-studio
• pytorch