A preliminary study assessing time-domain EMG features of classifying exercises in preventing falls in the elderly

Abstract

Falls are a common and an often serious problem in elderly people. The simple and effective way to prevent falls is regular exercises without requiring any weight at home. In order to promote the daily-life exercises in elderly people, the exercise recognition system based on surface electromyography (EMG) signals is presented. This research aimed to evaluate features extracted from four lower-limb EMG muscles during seven specific exercises in preventing falls. As a result, the suitable feature sets were identified that would provide an effective EMG pattern recognition. Eleven time-domain features were evaluated by using a statistical criterion method. Based on the information represented in EMG data, four features, consisting integrated EMG, waveform length, zero crossing and Willison amplitude, showed the best class separation performance of all studied features among four EMG muscles. A feature vector formed from such features is recommended to further improve the performance of the exercise recognition system in elderly people.