Features Selection for Human Activity Recognition in Telerehabilitation

Abstract

The article considers the problem of features selection for human activity recognition in telerehabilitation. The notion of telerehabilitation is considered. The types of physical activity that are necessary to conduct a telerehabilitation are indicated. A method for carrying it out using wearable devices is proposed. It is proposed to use a smartphone, which includes an accelerometer and a gyroscope to determine the physical activity performed by the patient. A literature review of approaches to the recognition of physical activity types is given. An experiment to select such features for the classification of physical activity that would allow for high classification accuracy with minimal computational costs, regardless of the location of the smartphone on the human body is carried out. As a means of carrying out the experiment, the MATLAB environment was used. For the experiment an open data sets has been used, which contains data from the sensors of the smartphone.