Optical Myography Using Convolutional Neural Networks for Estimating Finger Poses

Dies ist die aktuellste Version dieses Eintrags.

Kurzfassung

A plethora of techniques exist to rehabilitate an amputee’s lost limb. Active Hand Prosthesis (AHP) is one such tool which promises aid to an amputee to gain control over daily activities. Translating the user’s intent into appropriate movement of the prosthetic device, especially that of the hand is still a target to be attained by the various methods which try to acquire and interpret these biological signals. One such novel modality, known as Optical Myography (OMG) laid the proof of concept of mapping deformations on the surface of the forearm caused by muscle movement to estimate finger poses for an artificial hand. The surface movements were tracked using AprilTags stuck the surface of the forearm and by strapping the forearm to a frame in order to suppress external movement. Misdetection and missed detection of these tags can cause noise in the data acquired for the machine learning algorithm. This thesis aims to develop OMG for the estimation of finger poses by using computer vision to observe the muscle movements on the surface of the forearm thus obviating the need to rely on precisely detected tags. In order to do so, the machine learning algorithm used is a Convolutional Neural Network (CNN) trained on images of the forearm captured during the execution of the desired finger poses. Various feature extraction sources are studied before choosing the most practically applicable source to test on intact subjects.

Verfügbare Versionen dieses Eintrags

• Optical Myography Using Convolutional Neural Networks for Estimating Finger Poses. (deposited 23 Dez 2016 09:24) [Gegenwärtig angezeigt]