Transcript Powerpoint
3B8 Universal Design Innovation Group 14 Nitin Bansal, Felix Frank, Alan Myers, Tiarnan O'Kelly, Francis Yates The Problem Accidental pill ingestion cause significant problems: ”Less than 6% of the subjects knew about the toxic risks, sideeffects, or potential drug interactions.” Disordering of multiple pills is a regular occurrence for many elderly people Poor eyesight prevents many of the elderly sample group personally sorting pills PillAid totally eliminates this problem and will tell the user with verbal ques a pills identity Medicines can be of significance expense The ability to differentiate disordered pills eliminates the need to re-purchase pills Mission Statement PillAid aims to eradicate the serious health risk the elderly face due to dosage and pill identification errors. PillAid provides piece of mind by removing doubt about the identity of a pill and managing dosing and hence provide piece of mind for the user and loved ones. The Solution A device that scans the pills using neural network based technology Tells the user the identity of any pill and allows for dose tracking: Functionality Can be tought to recognise any pill Currently 4 different types of pill Currently can recognise up to 3 pills at the same time, this can be increased Can recognise mixed assortments of pills with high accuracy Could recognise an unlimited amount of different types of pills F1 Score: • Here we can see that extremely high F1 scores were achieved. This is a measure of the tests accuracy. • PillAid was trained on a small dataset of 3000 unique images and hence the accuracy achieved is a reflection of the devices good design. How it Works PillAid implements Machine Learning using a Convolutional Neural Network (CNN) PillAid is trained using sample data from any individual pill – approximately 10,000 training image per class are needed PillAid just needed to communicate over the internet using a simple python script interfacing with CNN on a server. More Detail A visualisation of the technology implemented: PillAid is currently trained on 7 classes and took approximately 4 hours to train on server based GPUs Demonstration Add video