Abstract

Gestures can be an interesting input method for various applications. There are several approaches on how to detect them, as new sensors provide different kinds of data to work with. One kind is muscle-specific data received via surface electromyography. It can be used to classify hand-gestures recorded by wearable sensors. In this thesis, the Myo armband and machine learning techniques are used to detect these gestures to play a game of Rock-Paper-Scissors. An interface for collecting sEMG data was developed in the Ubi-Interact framework. For gesture classification an existing Tensorflow model was trained and integrated into the JavaScript environment using Tensorflow.js. In the end, it was possible to build an application to play Rock-Paper-Scissors against a virtual opponent using hang-gesture input.

Results