This project is a real-time computer vision system designed to detect and classify recyclable materials using object detection. The goal is to support automated recycling and waste-sorting systems by identifying recyclable items on a conveyor belt in real-world conditions.
The system uses a YOLOv8 object detection model trained on the WaRP (Waste Recycling Plant) dataset. The trained model is optimized for deployment on edge devices, allowing real-time inference on low-power hardware such as a Raspberry Pi paired with an OAK-D camera.
The project demonstrates an end-to-end machine learning pipeline, from dataset selection and model training to deployment and real-time inference.
- Real-time object detection for recyclable materials
- Trained on an industry-relevant recycling dataset (WaRP)
- Optimized for edge deployment using ONNX
- Runs on Raspberry Pi with OAK-D depth camera
- Designed for conveyor belt and sorting facility environments
- Python
- YOLOv8
- OpenCV
- ONNX Runtime
- Raspberry Pi
- DepthAI / OAK-D Camera
This system is intended for use in recycling and waste management facilities where automated sorting can improve efficiency, reduce contamination, and increase recycling accuracy. The project focuses on practical performance rather than perfect accuracy, prioritizing smooth real-time operation.
models/– Trained and exported YOLO modelsscripts/– Inference and testing scriptsdata/– Dataset configuration and metadatautils/– Helper functions and preprocessing utilities
- Improve detection accuracy with additional training data
- Add material value estimation for detected recyclables
- Integrate tracking for item counting on conveyor belts
- Deploy with hardware accelerators (e.g., Coral USB TPU)
Kyle Jackson
Honours Bachelor of Artificial Intelligence – Durham College