Computer Vision Extended Perception System for Blind People

Abstract

There is a considerable number of visually impaired people in the world, who live their daily lives with limitations in their mobility due to the little information they have about their environment. This project proposes to use embedded systems to develop an assistance system for the blind, which, through machine learning models, describes the objects closest to the user and if they are known or unknown. To do this, two trained Machine Learning models will be used, the first to detect common objects and the second to identify particular objects. The intention of using both networks is to improve the accuracy of the system, by first detecting objects or people in an image and then classifying them with known labels (names or some identifier). The results obtained show the benefit of using these neural networks for the recognition of objects in the environment from a general database and then with a personalized one. Finally, to indicate the identified objects to blind people, the text of their labels is translated into speech.