An Intelligent Mobile Framework for OCR-Based Medicine Recognition and Voice-Assisted Medication Adherence

Abstract

Medication negligence is a major problem in long-term illness management. Nearly half of patients with long-term conditions do not take their medicines as prescribed. This problem is worse in low-resource settings, where poor label readability and limited pharmacist access make adherence harder. Most mobile reminder apps rely on manual data entry and generic alarms. They provide small help to the users who are unable to read small print or manage complex drug schedules. This paper presents a mobile health framework for medication management. An on-device text recognition (OCR), confidence-based quality checking, fuzzy medicine matching, and voice-assisted reminders are integrated based on the paper. Flutter, Firebase, and a three-layer Clean Architecture are used for building this system. An on-device OCR engine is used to read the medicine labels, and a weighted confidence score is used to check the recognition quality. It then matches the extracted text against an 11,825-record medicine directory using a six-rule fuzzy-matching engine. Reminders are delivered through native Android alarms paired with text-to-speech playback, supporting elderly and visually impaired users. Testing across six image conditions shows recognition reaching 98% success under favourable conditions, with predictable performance drops in poor conditions. A local caching layer cuts medicine-search time from 10–30 seconds to under 100 milliseconds. Reminder notifications reached 100% delivery reliability across 200 test alarms. These results show that combining lightweight on-device processing with accessible design can support faster and more reliable medicine identification and reminder delivery in resource-constrained settings; the framework’s effect on real- world medication-taking behaviour was not evaluated in this study and is identified as a direction for future clinical or user research.