One of the recycling issues is people are annoyed when they want to recycle an item they never met. Putting it into the wrong container does more damage to the Earth than just throwing it in the dumpster. Our goal is to help the citizens of Earth properly recycle their waste and find containers and recycling centers near them.
What it does
SkrapAI uses On-Device machine learning, powered by Firebase ML. Our goal is to help users classify the trash if it's recyclable or not and offer valuable and location-based information to users such as finding the nearest recycling center using data from OpenStreetMap near them and getting statistics about how much they contributed to saving the planet.
How we built it
The App can be separated into three parts. The Nearest recycling center part is fueled by OpenStreetMap data. The app communicates with a FastAPI server listening for GPS Coordinates and then later using a tool called Overpass which we query for anything with the tag
amenity=recycling and parsing it on the server to create friendly categories. Later we render that data using the Google Maps SDK
The model that delivers all this was built using Teachable Machine and it can also be trained with AutoML vision offered by Google Cloud. We used the following Kaggle dataset to train a model on 5500 images of Trash which took 4 hours of training.
Then there is the app which is built using the Flutter framework which allows us to build for both Android and iOS. Due to issues with the FirebaseML and Flutter Web, we couldn't deploy a Web version.
Challenges we ran into
Training a model in 24 hours without good hardware isn't easy. The amount of time that went into filtering images, removing classes, and playing with variables to get a model is time-consuming.
Accomplishments that we're proud of
We build a model to recognize Trash hooray! We also managed to make it accessible on phones and available for everyone to use all in 24 hours.
What we learned
What's next for SkrapAI
- Improve the model and add more categories
- Add even more meaningful from OpenStreetMap such as Opening and Closing Hours and contact information.
- Make it easier for users to search recycling centers and get directions