| The Gaussian mixture model (GMM), a modern statistical pattern recognition approach, has been shown to generally produce the best performance in speaker identification and verification. In this work, GMMs are applied to develop a limb motion classification scheme to discriminate multiple classes of limb motions using continuous myoelectric signals (MES). This GMM-based limb motion classification system demonstrates superior classification accuracy and results in a robust method of motion classification.;In an effort to optimize the configuration of this classification scheme, a complete experimental evaluation of the GMM is conducted on a 12 subject database. The experiments examine the GMM's algorithmic issues including the model order selection and variance limiting, the segmentation of the data, and various feature sets including time-domain (TD) features and autoregressive features with root mean square value (AR+RMS). The benefits of post-processing the results using a majority vote rule are demonstrated. The final GMM classification performance is compared to two commonly used classifiers: a linear discriminant analysis and a multilayer perceptron neural network. The GMM achieves 96.91% classification accuracy using a AR+RMS+TD feature set and attains 96.3% classification accuracy using a AR+RMS feature set for distinguishing six limb motions. It is shown to outperform the other motion modeling techniques on an identical six limb motion task. |
