Recycle Right

Inspiration

Only 27% of recyclable waste in Canada is actually recycled. That means 73% of things that CAN be recycled actually ends up being recycled! This is due to a lot of factors from lack of care, knowledge, contaminated recycling, and more.

In fact, 62% of Canadian respondents to a survey on recycling said they would recycle more if there were clearer information on how to. And 54% of respondents did not know they could recycle certain items, and 49% did not know how.

(https://www.wastedive.com/news/canadian-survey-identifies-confusion-as-a-significant-barrier-to-carton-rec/429033/)

The stats get even worse when looking at just plastic. In Canada 91% of plastic ISN'T being recycled. We wanted to improve our environment and keep our planet and country clean for not just us, but also for the future generations so we can up with Recycle Right. An automatic trash detection software to help improve these rates and keep our country green.

What it does

Recycle Right detects what kind of material is being shown and based on that determines whether or not the product is recyclable, compostable, shouldn't be thrown out (like batteries), or must be thrown out. This software could be coupled with smart trashcans or put at landfills to filter trash into compartments to reduce landfill waste and protect our environment.

How we built it

We used React, Tailwind, and JS for the frontend of the web app with Python + FastAPI for the backend. The detection software was YOLO26, used for classifying the different objects put onto the screen to determine whether it can be recycled or not.

Challenges we ran into

Classifying items into recyclable/non-recyclable, as well as their specific material was something we had to get over. Rate limiting + running out of credits was another issue we had to over came.

Accomplishments that we're proud of

Clear-cut explanation of why an item is recyclable/non-recyclable using Gemini API. Another accomplishment we are proud of is live interference and live tracking of objects.

What we learned

We learned to cope with API rate limiting and learn how to only make calls when necessary. We also learned how to optimize computer vision model use for CPUs, to ensure timely computation even without CUDA acceleration