A Physical Device to Help the Visually Impaired Read Money Using AI / Machine Learning in Third World Countries

Overview

There are over 285 million blind people in the world, with approximately 87% of them living in developing countries. However, in third world countries, there is currently very little technology to help the visually impaired, especially with financial independence. In this report we present the Machine Learning algorithms used to develop the device to help visually impaired distinguish between different forms of currency. Throughout this process, we tried the following approaches: the Google Vision API, Binary Classification, Filtering Data, KNN Feature Detection, and Transfer Learning. The final model used transfer learning, which had both the ability to work without the use of public WiFi and had a very high accuracy. The code for these various approaches can be found in the “References” section of the report. Using the various currency images, we formed a dataset that was able to test and identify the accuracy of these algorithms. Experimental results show over 94% accuracy with the Transfer Learning Model. The device is designed to be portable and hand-held. The device can distinguish between 1, 5, 10, and 20 dollar bills. Additionally, it can work in the absence of the Internet. Overall, the device is cost effective, portable and can be used in the absence of internet connectivity.

Competition history

• JSHS 2022 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.