Banknote Object Detection for the Visually Impaired using a CNN

Computer Vision (CV) is an area within the field of Artificial Intelligence (AI) which analyses images and video and attempts to identify and interpret the data contained in these. It aims to match or better the results a human could achieve given the same dataset. CV technology has major applications within the area of assistive technology. It has the potential to make the lives of disabled people easier by making the objects and systems they interact with more accessible. The aim of this project is to create a Convolutional Neural Network (CNN) suitable for use in a mobile bank note recognition application which can alleviate the struggle visually impaired people experience when trying to identify different bank note values. Limited previous studies have attempted to tackle this problem using a CNN. Additionally, these past studies have often neglected to include partial currency images in their datasets. This study uses data augmentation techniques to simulate partial currency images resembling those a blind or visually impaired person would take. The model created for this study achieved an average accuracy rate of 94% and was deemed suitable for use in a real-time currency recognition application.