Travel Footprint Estimator

Web interface of our app showing carbon footprint report based
Jupyter notebook we used to explore and analyze the data

Inspiration

We want raise awareness about how much impact travelling has on climate. And the problem we found is “How might me help big organizations to reduce their human footprint and raise awareness about how much impact the travel has?”.

Although there are few evident solutions already out there like the following:

Promote remote work among teams
Buy Smart Boards which facilitates meetings like Design Thinking Sessions

Nonetheless, these solutions have disadvantages, like:

It is much more better if you have your colleagues next to you instead on being on calls all day long
Although these Smart Boards enables multitouch collaboration and freestyle interaction, it has also a limit on the number of collaborators and it get can really expensive.

Our solution is the Travel Footprint Estimator.

What it does

Our application allows you to estimate your carbon footprint based on your localization data (Google Timeline) and predict you would if footprint shrink if you choose to switch to more ecological modes of transportation.

How we built it

Localization data from Google Timeline is analyzed by Python application and the reports are stored is PostgreSQL database.

We built web interface using Vue.js that lets you see the result of the analysis in the form of readable charts. Moreover, it let's you simulate the changes in your carbon footprint if you choose to switch to more ecological modes of transportation.

Backend for the web application was written Java with Spring Boot and Hibernate.

Everything is deployed to AWS.

Challenges we ran into

Originally we planned to use our own AI to classify GPS trajectories. It was supposed to be trained on the Geolife project data made available by Microsoft Research Asia. It turned out to be a difficult task since Google Timeline data is quite sparse (GPS localization is captured anywhere from every 30 second to every 5 minutes and not every few seconds as we hoped it would be).

We were also really hoping to prepare interactive plots for the prediction part, which turned out to be more difficult than expected.