Prospects of Wearable Inertial Sensors for Assessing Performance of Athletes Using Machine Learning Algorithms

Abstract

Understanding the area of science concerning the mechanics of human movement in sports through the use of wearable sensors is an emerging field. Although 3D motion sensors have been widely used to capturing human movements, onboard measurement has been a challenge. With advances in micromachining and sensor design, it has become possible to fabricate miniaturized inertial sensors for the onsite measurement of kinematics data. Many studies have involved the use of inertial sensors (comprising of an accelerometer, a gyroscope and a magnetometer) in sports. Despite inaccuracies owing to drift errors, inertial sensors have been potentially used to precisely measure hip angles in cycling, to provide a correlation between postural tremors and points scored in archery, to conduct real-time tracking of lower limb joints and torso kinematics of squat exercise, and to measure the grip angle and grip speed of the golf swing. The technical challenges associated with the accuracy of these measurements have been addressed in the literature using soft computing methods such as the Kalman filter, Gaussian smoothing for filter design, peak detection algorithms, the extended Kalman filter and quaternion. With advances in data sciences, new machine learning algorithms are being employed to study human motion using wearable inertial sensors, with possible applications in improving the performance of the sportsman while reducing the risks of injury (such as hamstring injuries, hip injuries, and rib stress fractures).