CNN Based Medical Assistive System for Visually Challenged to Identify Prescribed Medicines

Abstract

Developing assistive systems for visually challenged people is an active area of research in computer vision community. Such system provides a medical assistive tool to take the correct medicine at the right time as prescribed by the doctor and makes visually challenged people to live independently for their day to day activities. Many prototypes were developed to deal with misidentification of medicines but are incapable of determining exact pill picked by the person. This paper presents an automated system where feature extraction is done to recognize the pills based on structural, texture and Hu moments. If the pill is picked from the medicine box, the label present on the pill is considered for Text Recognition. Pill label and expiry date are extracted from the label and classified using Convolutional Neural Network (CNN) and is converted to speech. This audio is produced to indicate the person about the medicine picked. Experimental results proved that our system is better than existing works.