This project presents an AI-powered smart recycling bin that uses machine vision to automatically sort bottles and cans. A Raspberry Pi running a YOLO-based object detection model performs item classification, while an Arduino Mega controls sensors, actuators, and the conveyor system. Items move from a ramped intake onto a metered conveyor, where a camera captures images and servo-actuated gates route each item into the correct bin. The system aims to reduce recycling contamination by automating the sorting process in a compact, cost-effective design suitable for public spaces. The prototype targets 95% sorting accuracy with a processing time under 12 seconds per item, demonstrating the effectiveness of machine vision in automated waste management. This project is ongoing, with the next steps focused on building and integrating a physical prototype to validate system performance, mechanical reliability, and sorting accuracy.
Anti-pile panel prevents stacking and ensures single-item detection.
Clear camera view enables accurate classification.
Grouped slots increase spacing to keep items separated while allowing smooth belt motion.
Improves item isolation and consistent flow.
Gates open based on classification to direct items into correct bins.
Separate bins handle bottles, cans, and unclassified items.
The system uses a Raspberry Pi 5 (16 GB RAM) to run a YOLOv8s object detection model for real-time item classification, selected for its balance of accuracy and computational efficiency. A Raspberry Pi Camera Module 3 provides high-resolution image capture with full hardware compatibility. An Arduino Mega 2560 acts as the main controller, receiving classification results from the Pi and coordinating all electromechanical actions. Item tracking is handled by five high-speed IR break beam sensors, while conveyor motion is driven by two high-torque DC gear motors controlled via a TB6612FNG motor driver. Sorting gates are actuated using high-torque servo motors to withstand repeated impact loads.
Power is supplied from a 120 VAC source converted to 24 VDC and distributed through isolated DC–DC converters to separately power motors, servos, and control electronics. This architecture minimizes electrical noise, protects sensitive components, and ensures stable system operation.
A YOLO-based object detection model was trained using an online dataset of cans and bottles to evaluate its suitability for the recycling system. Training was performed on a laptop, where the model showed stronger detection performance for cans than bottles, likely due to dataset or shape differences. Although not yet integrated into the Raspberry Pi, the results demonstrated the model’s potential for real-time classification.