People who put recyclable stuff in the garbage bin when there’s literally a recycling bin right next to it. Why.

What it does

Automates the process of sorting recyclables when the trash is thrown. Takes in a photo and determines whether the disposable item is recyclable or not. Then sends a signal to the Smart Bin so that it can be sorted into the correct bin.

How we built it

  • Trained a model in Azure with tagged images of various types of garbage and recyclables.
  • The physical bin is made of scrap wood and metal found in our homes. The bin is separated into a trash side and a recyclable side.
  • The mechanical mechanism is a foam board attached to a DC motor controlled by Arduino which is connected to a camera to take a picture of the trash in question.
  • The image is processed by a python script and passed to the Azure Custom Vision API to identifies what type of material it most likely is. The value is returned which the Arduino responds by flipping the flap to expose the correct bin.
  • Created a basic website in React to show how an analytics dashboard might look like for multiple Smart Bins.

Challenges we ran into

  • Never used C#
  • Determining how to differentiate between recyclables and garbage
  • Optimizing the performance of our machine learning model
  • Communicating with the Arduino to send motor commands
  • Make the separator turn properly (a mechanical problem)
  • Using basic ReactJS to set up a sample data tracking website

Accomplishments that we're proud of

  • Successfully integrated Azure Computer Vision with a hardware component
  • It looks decent
  • The first time it recognized a Coke can was recyclable was pretty epic
  • It kinda works :)

What we learned

  • C#
  • Custom Vision in Microsoft Azure
  • Send C# data to an Arduino through Serial communication
  • Basic ReactJS
  • Wood is very hard to cut manually :(

What's next for Splash

  • Faster live detection.
  • Create our own dataset to better optimize model performance. Currently, the dataset we used has various angles, and getting pictures at the fixed angle of our camera would probably help.
  • Make this hack available for community use where recyclable restrictions are unique and customizable to community recycling standards.
  • Set up data collection for recycling bins. Currently, the website is simply a template, and it would be nice to feed data into it.
  • Improve the response system that gives a more clear indication whether the flap is actually set on the right side Make the power source solar rather than battery-based.
Share this project: