Assistive Device for the Visually Impaired Based on Computer Vision

Abstract

This paper proposed the design of an assistance system for the visually impaired based on computer vision. It includes an object detection tool powered by the YOLO v4 algorithm and Support Vector Machine and trained with a dataset of nine classes for detection and classification. Each class corresponds to items found at home extracted from the Open Images v4 database. A voice commands support provides a free-hand interface, and a voice guide allows the user to get feedback about the image detection and classification results. The system consists of an Android app, which implements Google Speech-To-Text and Text-To-Speech engines, and a mobile implementation of the YOLO v4 and Support Vector Machine algorithms. The results demonstrate that the application can be controlled by voice commands to detect objects in images with a confidence of 60% and a processing time of about 6.7 s.