A Multiaxial Data-Based Machine Learning Model for Exercise Motion Recognition

Abstract

Wearable devices using only accelerometers cannot correctly recognize the detailed motions of exercise performed by patients and cannot count the correct number of repetitions. Therefore, in this paper, we suggest a method to improve the recognition accuracy of detailed motions using machine learning technology simultaneously using an accelerometer, and magnetometer in a wearable device. We particularly improved the recognition accuracy of detailed exercise motions through machine learning using data measured on 9 axes. To verify its effectiveness, we conducted experiments for the recognition of each motion when the patients performed exercise consisting of nine motions with the 9-axis accelerometer attached to their wrists. Accuracy of 81.8% was recorded when only acceleration data was used, whereas accuracy of 91% was recorded in the 9- axis data obtained using three sensors, thereby improving accuracy by 9.2%.