AI-Based Pill Detection and Deblistering Spot Finder System for Smart Medication Dispenser

When on a medication regime, it is of extreme importance to take the right pills, and to make sure the medication is free of any contaminant. Current medication dispensers aim to aid patients; however, they still have inherent operational flaws. When dealing with blisters, the manual refilling of a dispenser is a lengthy and error-prone task. Also, as medication inside dispensers is stored out of its blisters it is exposed to air and humidity, which hinder its potency. To solve these issues, we aim to develop a new medication dispenser which removes pills from their packaging only when needed. This paper presents a system to be implemented in said dispenser, which automatically detects pills and locates their best deblistering spot, so they can be extracted in the most efficient way. The proposed system can be split into two phases: On the first phase a custom-trained Mask-RCNN detects pills and classifies them according to their type. On the second stage, a custom image processing algorithm calculates the ideal location where pills can be deblistered. Firstly, a dataset containing images of different types of pill blisters was created and thoroughly annotated for Mask-CNN. Afterwards, Mask-RCNN was trained to detect pills and classify them according to their shape. From the resulting pill masks and classifications, an image processing algorithm is applied to calculate the ideal pill deblistering location according to previous research. By correctly identifying each pill inside a blister and finding the best spot to extract it, the system can be implemented in a new medication dispenser, which only needs to be supplied with blisters.