Trash Code

Our cute logo :)
The web app's UI
Our tech stack!
Sample data pushed to MongoDB Atlus

Inspiration

Canadians produce more garbage per capita than any other country on earth, with the United States ranking third in the world. In fact, Canadians generate approximately 31 million tonnes of garbage a year. According to the Environmental Protection Agency, 75% of this waste is recyclable. Yet, only 30% of it is recycled. In order to increase this recycling rate and reduce our environmental impact, we were inspired to propose a solution through automating waste sorting.

What it does

Our vision takes control away from the user, and lets the machine do the thinking when it comes to waste disposal! By showing our app a type of waste through the webcam, we detect and classify the category of waste into either recyclable, compost, or landfill. From there, the appropriate compartment is opened to ensure that the right waste gets to the right place!

How we built it

Using TensorFlow and object detection, a python program analyzes the webcam image input and classifies the objects shown. The TensorFlow data is then collected and pushed to our MongoDB Atlas database via Google Cloud. For this project, we used machine learning and used a single shot detector model to maintain a balance between accuracy and speed. For the hardware, an Arduino 101 and a step motor were responsible for manipulating the position of the lid and opening the appropriate compartment.

Challenges we ran into

We had many issues with training our ML Models on Google Cloud, due to the meager resources provided by Google. Another issue we encountered was finding the right datasets, due to the novelty of our product. Due to these setbacks, we resorted to modifying a TensorFlow provided model.

Accomplishments that I'm proud of

We managed to work through difficulties and learned a lot during the process! We learned to connect TensorFlow, Arduino, MongoDB, and Express.js to create a synergistic project.

What's next for Trash Code

In the future, we aim to create a mobile app for improved accessibility and to create a fully customized trained ML model. We also hope to design a fully functional full-sized prototype with the Arduino.