Investigating Vision Based Sorting of Used Items

Abstract

This preliminary research investigates the development of vision-based methods for identifying objects in a sorting task. It addresses the issue of economic and environmental sustainability by exploring options to identify parts for reuse/recycling. Specifically, the problem of identifying electrical parts such as capacitors, voltage regulators, and potentiometers is investigated. The proposed solution will utilize a robotic arm and an overhead camera. A multiple object workspace scenario is investigated. A feature extraction algorithm identifies regions of interest and extracts data for classification. Three classes of objects are detected and classified by a backpropagation trained shallow neural network. An overall accuracy of 85.6% is currently achievable when tested in a real environment with new data.