Can Recycling TOMRA

Project description

How can the value of cans (deposit value) be determined without requiring them to retain their original shape?

Context

In the Netherlands, a new deposit system for cans has been developed, requiring cans to be returned to their original form for barcode recognition. While this approach ensures that the deposit value of cans can be easily determined, however there are challenges related to efficiency and environmental impact. Cans are often dented or deformed during use.

This project case by Tomra in collaboration with Fontys University, seeks to develop an innovative solution that allows cans to be recycled regardless of their shape. By using artificial intelligence (AI) and hardware interfacing technology, the aim is to design a system that is capable of recognizing the value of deformed cans, improving recycling process, and reducing the environmental impact of can disposal.

Results

AI models, specifically the YOLO algorithm, when trained on large number of data is able to differentiate between devalued and valued cans. We also trained a model to determine the shape of a can based on depth images and it proved to be possible. We also trained a machine learning model using data from microphone to differentiate between different states of cans. Therefore both image and sound recognition using AI is possible to determine the state of a can. Recent changes in technology allowed us to run the AI models on small devices such as the Jetson Nano.

Video