Recognition of daily human activities using accelerometer and sEMG signals

Abstract

Human activity recognition (HAR) is an important technology for ambient-assisted living, sport and fitness activities, and health care of elderly people. HAR is usually achieved in two steps: acquisition of body signals and classification of performed activities. This paper presents an investigation on the optimal setup for recognizing daily activities using a wearable system designed to acquire surface electromyography (sEMG) and accelerometer signals through wireless sensor nodes placed on the upper limbs of the human body. To evaluate the optimal number of accelerometer and sEMG signals for detecting the user’s activities, data recorded from a few subjects were used to train and test an automatic classifier for recognizing the type of exercise being performed. In this evaluation, that was performed on eight different exercises executed by four subjects, the automatic classifier achieved an overall accuracy ranging from 10.6% to 93.0% according to different selections and combinations of the signals acquired from the sensing nodes.