Inspiration
Our original inspiration came from our desire to incorporate almost all of the aspects of Computer Science and Data Science at Calvin University. As a team, we had a wide range of skills and interests, including IOT, Artificial Intelligence, Business Analytics, UX/UI, and Cloud Services.
What it does
Our application allows users to select their desired Calvin University building and locate trash cans via the specific layout map. Each trashcan has a "bar" to indicate its fullness level. Users also can see a variety of analytical graphs that correspond to a particular trash can or location. There is also a neural net of an LSTM in the background that forecasts what the expected trash levels are for a particular time.
How we built it
We used React JS to build our app and Plotly to show the analytics. The AI was built using Pytorch and Scikit Learn, utilizing the LSTM model. The database was hosted on AWS in a PostgreSQL management system. All data came from an ultrasonic sensor that was programmed to evaluate distance to closest detectable object. This was placed at the top of a trash can and as trash fills up the container, the distance decreases.
Challenges we ran into
The AI model took about 15 hours to train, fine tune, and build, so we were unable to truly connect it to our AWS server. We also didn't build 9 different sensors and since we were also building an AI based on time series data, we had to generate a lot of fake, but plausible data to build our model.
Accomplishments that we're proud of
We have a UI that displays Calvin maps and locations of trash cans on specific buildings can be found easily. The trash can bars also have color to indicate fullness, with red indicating that a new trash bag may be required, and green showing that it is relatively clean. We were also able to build a hardware device, attach it to AWS, and then build a web interface and AI model in one night.
What we learned
We learned that planning out a relational database diagram would've been helpful, both for our AI and for our original specs for our sensor.
What's next for Trash Predictions
We hope to add more floor plans for more of Calvin's buildings, and perhaps show more analytical data and reinforce the learning on our AI model with actual Calvin data and not generated data.
Log in or sign up for Devpost to join the conversation.