EAPR-Net: A lightweight framework for voice-assisted pill recognition in visually impaired users

Elderly individuals and visually impaired people are often at risk in managing their medication since they forget to take their medication, experience memory loss, deteriorating eyesight, and difficulty distinguishing between different pills. Misidentification of drugs can result in adverse health effects, and thus, accessible solutions are essential. To solve these problems, we propose a novel framework, called Enhanced Accessibility Pill Recognition Network (EAPR-Net), embedded in an Android mobile application. In addition to a CNN-based model, models have previously been proposed to overcome challenges like variability in lighting (through contrast enhancement), limits in power consumption with a network that is lighter than traditional CNNs for working on mobile devices, and alert feedback features with a voice assistant made possible through REST API. Users need only take a picture of their pills with a smartphone and receive immediate, audible, hands-off identification. The model was developed and tested on a synthetic dataset of 900 images from 14 common categories of pills, achieving a remarkable 98% test accuracy. The design of EAPR-Net not only ensures computational efficiency but also guarantees high classification accuracy for diverse real applications. By combining advanced computer vision methodologies with user-friendly accessibility features, this framework advances medication safety to a new level. Finally, EAPR-Net enables the elderly and visually impaired to become self-sufficient in prescription management and less dependent on caregivers, resulting in an overall better quality of life.