Recognition system of finger motion pattern based on AR model coefficient estimation

A new type of finger articulation angle recognition system based on surface electromyography (sEMG) signals were established. Especially, velocities of fingers can be effectively predicted with a few sEMG channels. This paper mainly collected the sEMG signals during forefinger motions which are slow, medium and fast velocities, respectively. A new method were proposed which acquired more accurate estimate of the velocity variations of the finger. Firstly, fast Independent Component Analysis (FastICA) algorithm based on the largest negative entropy was used to predict the acquired signals, and then separate effective operation of the signals. Then autoregressive (AR) parameter model U-C algorithm was used to extract characteristic coefficient. Finally, a RBF neural network was designed, and the input is computational AR characteristic coefficient, the outputs are three different velocities of fingers motion. The experiment data were collected from healthy subjects' four muscles including Index finger extensor (IFE), Middle finger extensor (MFE), Palmaris Longus (PL) and the flexor carpi (FC). RBF neural network was used as a classifier to classification and recognition different finger movement action, and a satisfactory results has achieved.