A camera-based garbage bin that sorts recyclable from non-recyclable waste on its own — image processing and machine learning replace the unreliable sensors most smart bins rely on. Published at ICMIME 2024.
Most "smart bin" research sorts waste using sensors — moisture, capacitive, metal — which are fragile and prone to failure from dust, heat, and noise, and few of them target the specific problem of recyclable vs. non-recyclable waste.
We built the alternative: a camera and a trained machine learning model doing the sorting instead of a sensor guessing at it, with a servo motor acting on the result. Built with a team of RUET engineering classmates and published as peer-reviewed research at ICMIME 2024.
ESP32-CAM captures and streams real-time images.
A model trained on recyclable/non-recyclable product images makes the call.
Anticlockwise for recyclable, clockwise for non-recyclable.
Avoids the fragility of moisture/capacitive/metal sensors.
Minimal delay from detection to disposal.
ESP32-CAM, FTDI programming, a Vero-board circuit, and a PVC-housed dual-bin enclosure.
In testing, the system classified waste as recyclable or non-recyclable with ~90% accuracy, processing images in real time with minimal delay. The servo motor reliably routed items to the correct bin without human intervention, cutting contamination between waste streams.
Accuracy dropped in low-light conditions, and scaling to more varied urban waste profiles would need a larger training set — both noted as directions for future work.