Vision-based Segmentation of Continuous Mechanomyographic Grasping Sequences for Training Multifunction Prostheses

In designing mechanomyographic (MMG) signal classifiers for prosthetic control, the acquisition of long, continuous streams of MMG signals is typically preferred over the painstaking collection of individual, isolated contractions. However, a major challenge with continuous collection is the subsequent separation of the MMG data stream into segments representing individual contractions. This paper proposes an automatic, vision-based segmentation method for continuously recorded MMG data streams. MMG data acquisition was synchronized with transverse plane video acquisition of functional grip sequences. The automatic segmentation system can track a hand, recognize grips and detect grip transition times as well as extraneous hand movements. The system recognizes two grips with an average accuracy of 97.8plusmn4%, and seven grips with an accuracy of 73plusmn20%. The contraction initiation and termination times agree closely (within 1.3plusmn1 frames) with values obtained manually