Poubelle

First thing you see when you open Poubelle
Zoomed in view of datapoints, to demonstrate granularity
Trash can view

Inspiration

When we got to pennapps on Friday, we didn't do as much programming as we did touring. We took walks around the campus and furthermore around the city. We noticed that there were some places that tended to accumulate trash more than others. In the bigger picture, we realized this problem extends to any medium to large sized metropolitan city. Thats when we started to work on Poubelle.

What it does

Poubelle is a live heatmap of trash hotspots in Philadelphia. We used municipal surveys and data sets to populate the initial state of Poubelle, however adjacent to the main heatmap, we built an IOS app that lets a pedestrian user take a picture of a location, have an AI determine how much trash there is, and add the geo-coordinates to Poubelle's heatmap. The data outcomes span various purposes: A tourist will check Poubelle to avoid more trash filled areas, street sweepers will check Poubelle to optimize their cleaning efforts, and city officials will check Poubelle to monitor and improve waste disposal efforts -- such as optimal locations for trash cans. On that last use case, we also show a city-wide map of all current trashcans and allow for users to place their own at custom locations to see how it would impact the trash distribution in the city given the existing patterns and conditions.

How we built it

We built Poubelle using the following technologies:

nextjs
AWS
mask r-cnn
Tensorflow
mapbox
IOS/ Swift-UI
Firebase First we built a rest API around the Municipal Data, then we expanded on the UI to make it functional, and then after that we made a user centric IOS app for people to submit potential trash hotspots.

Challenges we ran into

Initially, figuring out how to populate Poubelle with existing trash hotspots was challenging. We though we may have to go out ourselves and do it manually, or scrape Google Street View, which was not really practical. Instead we were fortunate enough to find municipal database.

Accomplishments that we're proud of

In the home stretch of development, we made an IOS app that was a key feature in making Poubelle the user contribution based platform that enables it to be so effective. It was a simple but powerful tool that greatly increased the dimensions of our project, enabling for user contribution.

What we learned

We learned that scraping Google Street View is extremely difficult for a production environment especially when you need to orient the camera at a certain angle. Also, we learned how to deploy trash detecting neural networks in production using a REST API. Finally, we learned how to make a website that can process submissions from the mobile app in real time.

What's next for Poubelle

We’d like to refine our trash can simulation to use a bayesian model with data from google street view. We’d also like to use the google street view data in order to add more granularity to the municipal datasets such as how much litter a particular location has and the type of litter. Also, once the google street view analysis model is refined, we will be able to run simulations for locations without municipal city data.