ReciclAmBarna

glassbottle
can
plasticbottle
glass
plastic
unkown
arquitectura
dataset
features
validation

Inspiration

Barcelona is a highly touristic city where recycling can be confusing due to different systems across countries. Inspired by this issue, we built a smart assistant that helps users identify where to dispose of waste correctly in real time.

What it does

This is an interactive system that helps anyone recycle correctly in real time.

How we built it

We combined computer vision and hardware:

A model trained with Edge Impulse classifies objects (plastic, glass, others)
A Python backend runs the model and processes predictions
An Arduino controls LEDs to indicate the correct recycling bin
A simple web interface displays detections

Challenges we ran into

The Arduino Uno Q cannot run machine learning models due to its limited memory and processing power. Additionally, it cannot interface directly with a USB webcam, since webcams require a host system (like a PC or a Raspberry Pi) to operate. For this reason, we moved the computer vision inference to an external process.

Accomplishments that we're proud of

Built a working end-to-end prototype combining computer vision and hardware
Deployed a custom-trained model using Edge Impulse
Designed a system that works in real time
Overcame limitations of Arduino App Lab by redesigning the architecture

What we learned

Microcontrollers like Arduino Uno Q are not suitable for running complex ML models
Separating AI processing from hardware control improves flexibility and reliability
Rapid prototyping requires prioritizing what actually works over ideal solutions
Used SSH and SFTP to access the host device running the backend