Técnicas de reconhecimento de imagem para incorporação em ferramentas de auxílio a deficientes visuais

Abstract

Visual impairment can cause many limitations to a person, which includes working in basic home duties, professional activities, and also changing the way the world is perceived by him or her. The advent of new technologies allows the design and exploration of new kinds of aid devices for the visually impaired. Image recognition techniques have long been used as a tool to compensate or complement human visual limitations. More recently, advances in machine learning techniques, such as deep learning, have offered new possibilities to build aid tools devices that can play a significant role in reducing the limitations suffered by blind people. This work presents the results related to the development of an image recognition module to be used as part of an existing embedded aid tool for visually impaired people. The module is based on convolutional neural networks, and is initially targeted to the recognition of the state of indoor home objects.