Creating A Smart Pill Dispenser with AI Technology to Help Fight the Medication Nonadherence Problem With Raspberry Pi Technology

The Project

Medication nonadherence is a critical problem within special needs groups, such as elderly patients living independently or people struggling with mental health. The leading cause for medication nonadherence is forgetfulness and procrastination. The project's objective is to leverage AI and internet technology through the Raspberry Pi 4B and Raspberry Pi camera to solve the problem in a way that is reliable, convenient, and affordable. This project has three main components: a physical device, an image classification AI, and a web server. Together, it evaluates whether the medication was taken and conveys the result to the caregiver on the website. The prototype was successfully tested with 100% accuracy and >99% inferencing confidence. The website was able to present the medication history data collected correctly. With an operating cost of $4/year, this prototype meets the requirements of being an accurate, cost-effective, and easy-to-use medical device to address the medication nonadherence problem.

View this project on vimeo

Team Comments

Between my grandmother and her friends, they frequently forget to take her medication, including critical heart medication. I researched this issue to understand the magnitude of this problem in society. It causes an estimated 125,000 deaths and 25% of hospitalizations in the United States annually.

Since my project involves many things from 3D printing to physical design involving the Raspberry Pi and circuits to AI, it required a lot of learning. The amount of time conflicted with school, and I eventually made it my goal to 30 minutes everyday, from watching Youtube videos to building it.

One of the factors preventing this being a retail product would be the size. The container is tall, since it used a Raspberry Pi camera that had a minimum focus distance. Although I made modifications to lower the focus distance, it was still large and in the future, I’d use an Arducam macro lens.