Please use this identifier to cite or link to this item:
Title: Real Time Surrounding Identification for Visually Impaired using Deep Learning Technique
Authors: Gupta, Hardik
Dahiya, Dhruv
Dutta, Malay Kishore
Travieso, Carlos M. 
Vásquez Núñez, Jose Luis 
UNESCO Clasification: 3307 Tecnología electrónica
Keywords: Deep Learning
Object Detection
Real Time
Tensor Flow
Visually Impaired
Issue Date: 2019
Journal: IWOBI 2019 - Ieee International Work Conference On Bioinspired Intelligence, Proceedings
Conference: 2019 IEEE International Work Conference on Bioinspired Intelligence, IWOBI 2019 
Abstract: Navigating around unfamiliar places and performing other day to day physical tasks are some of the biggest challenges faced by visually impaired people. It is extremely difficult for visually impaired people to commute or perform daily tasks without physical assistance. The conventional methods to aid visually impaired people mostly uses sensors to estimate distances from objects which is very inefficient, expensive and difficult to use without assistance. The proposed work presents a way to provide sight to visually impaired in real time using deep learning by identifying some familiar places used in day to day life like Restrooms, Pharmacies and Metro Stations. This method uses convolutional neural networks to identify signs of public places which are similar around the globe. The proposed work was tested on large varying database and achieved a high accuracy of 90.992 percent. The experimental results show that this method for identifying Restrooms, Pharmacies and Metro Station signs is efficient, has low computational time and fulfils the needs of visually impaired people up to a large extent.
ISBN: 9781728109671
DOI: 10.1109/IWOBI47054.2019.9114475
Source: IWOBI 2019 - IEEE International Work Conference on Bioinspired Intelligence, Proceedings[EISSN ], p. 41-44, (Julio 2019)
Appears in Collections:Actas de congresos
Show full item record

Page view(s)

checked on Sep 20, 2020

Google ScholarTM




Export metadata

Items in accedaCRIS are protected by copyright, with all rights reserved, unless otherwise indicated.