iPark

Web-App

Inspiration

What it does

It uses trained time-series tensorflow models to predict real-time free parking spaces in the city of Zurich.

How I built it

We used tensorflow and keras to train time-series dependent data to predict multiple points in the future. We used the Tom Tom API to load maps and connected the backend-frontend using node.js and json!

Challenges I ran into

Connection of Front-end and Back-end. All the teams are proficient in Deep learning, NLP and computer vision but not so much with web development. There were some out-of-serive parking spots (0s in the data for a long time) we had to manually remove them to avoid dataset bias.

Accomplishments that I'm proud of

The models trained have very high accuracy and can predict up to 72 points in the future (around next ~12 hours) for all parking locations (Currently only 4 models)

What I learned

LSTMs models (just 4 layers) are pretty good at learning time-series data.

What's next for iPark

Adding a Parking wallet where a person can gain parking points by choosing the most sustainble parking option. He/She can collect and redeem these points for some parking space.