TrashAI

Video Link

https://imgur.com/a/Q3IqCAh
Presentation: https://docs.google.com/presentation/d/1xwvs9L-JuEliBrQKbKuWLpo5DtARMEbbf7r-RdLlHXA/edit?usp=sharing

Inspiration

Any UW student has witnessed the mess left on top of the waste bins after lunch hours at the HUB. Human error and uncertainty results in haphazard sorting of trash, with the compost recycling and trash all filled with everything almost randomly. This is not a unique problem to the HUB, many of us are acquainted with the process of leaving a restaurant and having to sort through what belongs where. Many establishments have made great progress in using compostable packaging, but this progress means nothing if the packaging never makes it to the compost! We sought a solution to this problem that will make sorting trash easier and more accurate. This both improves the customer experience and benefits our environment. This is what our smart trash can seeks to do.

What it does

The EcoBin is able to automatically detect the classification of waste items, and sort them into the correct bins for us.

How we built it

TrashAI uses a webcam connected to a Raspberry Pi to take pictures of items as they enter the bin, and our CNN Tensorflow model then classifies them and decides if they are trash, compost, glass, plastic, metal, cardboard, or paper. The Pi sends the decision to an Arduino which then angles the flap to drop items into the proper bin.

The Tensorflow model was generated using transfer-learning on top of VGG16, a 16 layer ImageNet model. More information and the model can be found on the first github link

We have a management dashboard built on HTML/CSS/JS that has live data of the items in the bin through Google Firebase.

Challenges we ran into

Our largest challenge was getting all of the separate components of our project to work together in tandem.

Accomplishments that we're proud of

We are proud of our TrashAI/EcoBin being able to sort between trash and recycling as accurately as it does.

What we learned

We learned about the challenge of integrating multiple different software/hardware components of a project.

What's next for TrashAI

In the future, TrashAI will be able to sort for up to seven different types of materials as well as trash and recycling. Additionally, the AI EcoBin uses to identify waste items can be trained better for even more accurate classification.