San Diego being a coastal town, we have our fair share of issues with litter. We thought we could help alleviate the issue using drones and machine learning.
What it does
Trash Hero is a proposal for a drone system that uses computer vision and neural networks to recognize litter, and then collect and dispose of it, with no need for human intervention.
How we built it
The computer vision component uses Google Cloud Vision, however if we had more time we'd have wanted to use OpenCV/TensorFlow/Keras. If we had access to the appropriate hardware and necessary time, we would have used a microcomputer such as an Arduino with a camera shield for the drone itself,
Challenges we ran into
We had to teach ourselves about neural networks and how to work with APIs. Also, we had limited computing resources which necessitated us using a prebuilt neural network model - Google Cloud Vision. We had trouble initially integrating Google Cloud Vision into Python, and the hardware was not available on-site to make a drone, only the software. We also had issues getting camera input working.
Accomplishments that we're proud of
Getting Google Cloud Vision working was a really cool experience.
What we learned
We learned about neural networks and AI, and about API programming. *
What's next for Trash Hero aerial drone
Switch to OpenCV/Keras, get camera input working, get the software working on a drone platform instead of a computer.