EMG Hand-Gesture Recognition for Prosthetic Control

A myoelectric prosthetic hand is controlled by recognising the user's intended gesture from surface EMG recorded on the forearm. This project uses the public 8-channel armband recordings of 36 people (UCI EMG data for gestures), extracts the classic Hudgins time-domain features from 200 ms windows and compares linear discriminant analysis (the clinical standard), a support vector machine and a 1-D convolutional neural network on the raw signal - tested on people the classifier has never seen, which is what a new prosthesis user experiences. It also shows how much a random window split overstates accuracy, how a few seconds of personal calibration help, how fast the controller reacts with a majority vote, and which forearm muscles explain the confusions (OpenSim wrist model). A web application scores recordings, calibrates patients, plays the gesture stream back as a moving hand and raises control alerts, on five databases. A GNU Octave cross-check and a complete MATLAB version are included.

Category: Biomedical Engineering

Biomedical Engineering Project BM02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.