Magnify Trash

Inspiration

Recycling and waste management is confusing. There were many times when we were out in public and spend way too long trying to decide which bin to throw our trash in, and simply end up throwing our item in the landfill bin. On top of that, recycling laws differ from city to city, county to county, making the whole process complicated and places a lot of cognitive load on the user.

As members of Generation Z who hyper fixate on the future, we have come to realize the profound impact that small, everyday behaviors can have when they are compounded over time. According to the United States Environmental Protection Agency (EPA), out of the 146 million tons of municipal solid waste (MSW), only a small 32% was properly recycled. Additionally, our own user research states that out of all participants 50% throw away garbage more than twice a day, but only 20% believe that environmental issues are relevant in comparison to other societal issues.

Based on this data, our team aims to build a solution that simplify waste management to enhance recycling efforts and promote proper recycling habits to help build a more sustainable future for not only our generation, but also future generations to come.

What it does

Magnify Trash is a hardware-software solution that uses image recognition and sensor technology to detect the type of trash in 4 clades (can, paper, plastic and trash), and opens the lid to the respective bin for the user to deposit. It also detects whether or not the current bin is empty. Afterwards, the hardware sends the data stream asynchronously to the cloud which gets updated in a cross-platform mobile application. Inside the mobile application, users of all ages experience rewards based on certain number of items recycled milestones. This gamification enables motivation and empowerment towards recycling items without much overhead in time spent worrying about which correct bin to deposit waste inside.

How we built it

Our initial prototype consists of a cardboard box with microcontrollers, drivers (LN2003), and IRSensors. The IRSensors detect whether the trash can is empty or not and we also established a bluetooth connection through Universal Asynchronous Receiving Transmitting (UArt). The microcontrollers transmit the corresponding character arrays of data, which tell the bins are full or not full. While trash goes in, we implemented a custom machine-learning (ML) model acting as a utility of the prototype and an external camera application that predicatively assigns our 6 clades as 1 or 0, where 1 is the most likely, and 0 being most unlikely. After that, all the data from the ML model is sent to the cloud and then received through asynchronous communication between the Firestore server and frontend to update the individual user metrics.

Challenges we ran into

The AI model stretched us in a myriad number of ways. Notably, we decided to create and train a custom ML model, which required a great deal of computational power. The sheer amount of 25k+ images of different types of potential recyclables coupled with the model’s complexity, slowed us down for almost the entire hackathon. Additionally, since the model relies on a camera to classify and recognize images, we experienced variations in lighting and perspective. Additionally, uploading the data stream to the cloud and receiving that proved strenuous due to the variable amount of latency and reliability, creating feelings of uncertainty.

Accomplishments that we're proud of

Our team learned the advanced capabilities of UArt bluetooth and quickly applied the new concepts as a team to connect the hardware to the AI model. In addition, our team feels grateful and proud of creating a custom machine learning model from scratch to accompany the challenging hardware components of synchronization and utility support. We are also particularly proud of everyone lifting each other up as we combined our unique strengths together in harmony.

What we learned

Overall, our team learned about a deeper knowledge of hardware architecture, cloud storage, and creating custom AI models. We build the hardware using the industry standard microcontroller ArmcortexM4 for the first time during this 36 hours. It was also our first time working with image detection AI models. On top of that, we also learned about patience, resilience, and the concept of “eat your vegetables”. This means trying out new concepts and experiences, as mentioned in the learning section, that may seem hard at first, but are worth it in the end.

What's next for Magnify Trash

The movement for fighting against the dangers of waste mismanagement extends way beyond 36 hours. Our team potentially plans on publishing Magnify Trash to the App Store and GooglePlay store while finding a sustainable manufacturer to produce patented versions of the bins. Distributing the trash cans globally across the world would lower the growth towards a wasteful future and instead, move us forward in a more sustainable manner.