Vocal Vision for Visually Impaired People

EasyChair Preprint 9927

Abstract

The project is an endeavour to provide the visually impaired with a friend, more dependable and dedicated than most humans can ever be. There are lots of services for the blind to find the objects, but none of them manage to give an avoid description of the same. It present a unique way of being able to navigate with assistant from the only AI a blind person will ever need. Object Detection Using sophisticated object detection algorithms, implemented from scratch on a tensorflow library on a native android device by incorporating a trained classifier,  detect the object a user wants to find within an environment. It built a cloud API through which the app fetches further information about the object as caught in the picture. A reinforcement learning based AI model is to be incorporated to make the predictions better with time while training. Using Neural Network based classifier which helps to choose the right direction to take at every frame as recorded from the back camera of an android device. User can use dictionary which is also voice-operated and artificial intelligence based that can translate much better even in a full sentence. Detecting objects is the most significant part of our project. Object detection helps blind user to know about their current environment. Multi-Class Object Detection on Mobile Live-Video Stream using Deep Learning Convnets, to assist the Visually Challenged, or to signal an incoming threat.

Keyphrases: AI, Blind people, deep learning, neural network, object detection

@booklet{EasyChair:9927,
  author    = {Esther Sharon I and V.V Gomathi},
  title     = {Vocal Vision for Visually Impaired People},
  howpublished = {EasyChair Preprint 9927},
  year      = {EasyChair, 2023}}