**** LIVE DEMO: We are proud to present a live demo of our app on https://d1tdd4epdav6jy.cloudfront.net ****** _ Should be fully functional besides the unlock function of the car. Sorry we cannot let everyone unlock the demo car _


We started with the insight that there is only one quick way to improve alertness of a tired driver - doing some kind of physical exercise during a stopover. The idea of unlocking the car by doing some kind of exercise - squats as a proof of concept - was born. When we were talking to potential users of our product we found out that many people love to add some activity to their every day live - especially when they spent a sizeable part of it in their cars.

What it does

Fit 2 Drive utilizes pose estimation via deep learning to track different body motions of the user who performs one of our exercise plans. Our software then automatically opens the car through an API call once the user has finished his exercise. We create a coherent user experience around our core functionality by providing gamification (challenges, experience points and badges) and social features (share your achievements on facebook).

How I built it

The image recognition algorithms runs on a Javascript tensorflow implementation directly in the browser of the user. The UI around it is built in React.JS with Redux as a data store and Webpack and Sass as build tools. Our App uses the html5 camera api to access the device's camera stream

Challenges I ran into

While we were trying out different implementations of pose estimation algorithms it became soon evident that sending the video stream across the internet in order to process it on our servers would be a very complicated process - so we had to look for a deep learning framework that works across multiple devices, has good community support and is in a mature development phase.

Also, setting up the React/Redux architecture with user data management, local routing and templating as well as developing a appealing UI design was by far not a trivial task

Accomplishments that I'm proud of

All our image processing is happening locally on the users device - that means (1) no video or pictures of our users are ever send across the internet and (2) the app fully works offline. Scalability is also less of an issue because we do not need server infrastructure like on a alternative architecture with backend image processing.

What I learned

Tensorflow.js is a nice and robust implementation, opening a car via API is a cool thing and we saw again dozens of caveats when working with React/Redux/Webpack/Sass

What's next for Fit 2 Drive

In term of product development we think there is only one way for Fit 2 Drive to unlock its full potential - get in in the hands of our users, listen to their feedback and develop it as close to the market as possible. While we are a end-user product, health insurance providers may be willing to add it to their existing programs supporting gym memberships and routine screenings. Also, rental car companies and the forwarding industry are interested to prevent accidents due to microsleep and to provide this service as an additional perk for their customers/employees

Built With

Share this project: