CASPIAN JOURNAL

MANAGEMENT AND HIGH TECHNOLOGIES

Facial gesture recognition by electromyography’s signal

Read Starchenko I.B., Budko R.Yu. Facial gesture recognition by electromyography’s signal // Caspian journal : management and high technologies. — 2016. — №1. — pp. 39-50.

Starchenko I.B. - D.Sc. (Engineering), Professor, Southern Federal University, 2 Shevchenko St., Taganrog, 347922, Russian Federation, star@fep.tti.sfedu.ru

Budko R.Yu. - student, Southern Federal University, 2 Shevchenko St., Taganrog, 347922, Russian Federation, budko@yandex.ru

The article presents the results of an experiment on the facial muscles electromyographic signal processing (EMG), based on an algorithm of radial basis function neural networks (NN). For the training of the NN proposed to use the EMG signal characteristics in the time domain obtained by the non-overlapping windows. Facial EMG was recorded with a group of subjects. As part of the pre-processing of the signal used in the procedure which provided noise reduction, filtering, smoothing, segmentation, dimension reduction, feature extraction. Study and compare the efficiency of use as input for training NN nine signs EMG learned as a function of time: the integrated EMG; average; the average unit value; finite difference; the sum of elementary areas; dispersion; standard deviation; signal length; the maximum peak value. Evaluating the effectiveness of the use of these signs it was carried out on two most important parameters to be used in applications in real-time: NN performance and training time. Best performance for NN result was obtained for the characteristic В«Maximum Peak EMGВ» (recognition accuracy of 93,4 % on average for all subjects) at high speed training (0,25 seconds). A comparison of results of the proposed algorithm with the method of support vector machines and multilayer perceptrons NN. Proven high performance algorithm, radial basis function. The resulting algorithm and neural network based on it can be used in problems of constructing man-machine interface in real time (for example, to control the wheelchair).

Key words: электромиограмма, мимические движения, распознавание, обработка сигнала, искусственные нейронные сети, извлечение признаков, радиальная базисная функция нейронной сети, electromyography, facial movements, recognition, signal processing, artificial neural