ABSTRACT

The visually impaired are limited in resources that allow them to be independent in their daily lives. Specifically in restaurants, the visually impaired face problems with reading a menu, visually confirming if the served food has been served to-order, and navigating between objects on the table. Computer vision and AI can be used to address these problems. The goal of this research was to investigate the technology of computer vision and machine learning and to utilize it’s capabilities to assist the visually impaired. Our objectives were to explore the fundamental machine learning features and capabilities so that we could create a deep learning neural network capable of classifying images of foods, evaluate its performance, and improve upon it to increase its accuracy past 90%. In addition, we wanted to understand the hardware and how our model could be deployed through a device. We investigated self-defined neural networks that yielded low accuracy and shifted to transfer learning with pre-trained network models which were able to accurately classify images in our custom dataset to around 92%.