AI Assistance for Visually Challenged People

Author: Pranav Jadhav, Student, Department of Computer Science, Vishwakarma University, Pune, Maharashtra, India

Published On: 2024-04-22

This research paper serves the AI/ML model which measures the distance of specific objects from the host without any additional sensors other than the camera’s own sensors. It also detects the Indian denominations and report them through hearing aid. Model is made with CNN and OpenCV which integrates the computer vision concepts. Custom models are built which are exported to pkl file for runtime access. We use around 1500 images of Indian notes and augmented them for further processing. The model predicts the denomination from real time video and gives the audio response. It classifies notes with a 92 percent accuracy and is much better for a real time video input. This model can be directly implement for assisting visually challenged people. The model handles the error with 0.4 adjust score. It even works best with just 15fps camera. This makes it cost effective and easily scalable

Key Words:  Visually Challenged, Safety Research, Work Distance, Measurement Notes Classification, Video Application