Toward Markerless, Noncontact Optical Myography for Prosthetic Control of Multiple Degrees of Freedom

Accurate hand gestures and movement intent detection are crucial for the intuitive control of prosthetic devices. Commercially available technologies such as surface electromyography (sEMG) require extensive training and direct contact with the residual limb to detect motor intent. Recently, optical myography (OMG) has shown promise for hand gesture recognition. When compared to conventional approaches, OMG uses a camera with computer vision techniques to observe morphological changes on the forearm. This article presents a fully markerless approach for hand gesture recognition with a simple statistical feature by looking at the forearm. We systematically analyze the factors affecting classification performance. We achieved approximately 94% average accuracy with nondisabled participants for five gestures. We further implemented our strategy on two individuals with egocentric views and achieved 72.4% and 77.4% mean cross-validation accuracy with the decision tree classifier on two trans-radial amputees, respectively.