Note: this project won Top 5 at SB Hacks VI (badge only presented to 1st & 2nd)
A few days ago, I came up to a traffic intersection as a pedestrian. I waited a couple seconds, motioning for the driver to go, as he had arrived first. We then came to that awkward standstill where both person wants to be polite and let the other person go ahead. Eventually, I gave in and crossed the street. The event got me thinking though, and after doing a lot of research into how ineffective stop signs are, I decided it was a problem worth pursuing.
What it does
StopLess is a mobile app that uses AR & CV to transform regular, dangerous stop sign intersections into interactive, safer intersections kept in sync on the Cloud. It currently recognizes when a stop sign is visible, checks for pedestrians who are nearby, and then logically decides who gets to go first.
How I built it
Computer Vision originally using GCP Vision, but then switched to Flask + TensorFlow due to latency. Augmented Reality using React Native SVG. Mobile App using React Native + Expo. Cloud communication using Firebase Firestore snapshots.
Challenges I ran into
I spent a lot of time trying to get a TensorFlow model running in the React Native app, but it's near impossible without ejecting from Expo. So I settled on hosting the ML model on localhost which is decently quick.
Accomplishments that I'm proud of
Building the entire stack as a single person. I usually have a full team where everyone focuses on one thing, and although I've worked with every piece of the stack on different projects, I've pretty much never had the opportunity to work with the entire stack on a single project.
What I learned
I learned that hackathon projects are actually feasibly completable by a single person while also having a decent amount of time to enjoy oneself.
What's next for StopLess
Just like Google Maps on the Smartphone revolutionized navigation, StopLess on AR Goggles could revolutionize traffic control. Stop Signs are just the start of this revolution; Traffic Lights are next. In fact, a single Traffic Light costs $600/yr in electricity. Countless developing countries often lack traffic infrastructure for a number of reasons, resulting in extremely unsafe intersections and economic losses in productivity. Virtualizing traffic lights also enables them to be connected to the Internet of Things, and be modified using Machine Learning on the huge amounts of data produced by car transport.