Myoelectric pattern recognition using dynamic motions with limb position changes

Myoelectric pattern recognition discriminates myoelectric signal patterns to recognize human gestures because myoelectric signal exhibits muscular activities of human motions. This technique has been broadly researched for prosthetic devices and human computer interface. However, it still has robustness issues caused by various conditions: limb position changes, long-term uses, electrode shifts, and skin condition changes. These issues have compromised the reliability of pattern recognition technique in myoelectric systems. In order to increase the reliability in the limb position effect, this paper develops a myoelectric system using dynamic motions. For the classification of dynamic motions, dynamic time warping technique was used for the alignment of two different time-length motions and correlation coefficient was then calculated as a similarity metric. For comparison purposes, static motions were classified by Naive Bayesian classifier for the classification and Fisher's Linear Discriminant Analysis for the dimensionality reduction. Static and dynamic motions were collected at four different limb positions for estimating the robustness to the limb position effect. The statistical analysis, t-test ( p<0.05), showed that dynamic motions were more robust to the limb position effect than static motions from all eight subjects when training and validation sets were extracted from different limb positions. Subject 5 showed the best classification accuracy: 97.59% with 3.54% standard deviation (SD) for dynamic motions and 71.85% with 12.62% SD for static motions when training and validation sets were picked from different limb positions.