Abstract:

In this study, a classification study of human lower extremity activities was carried out using a wearable miniature inertial sensor. The inertial sensor used in the study has nine degrees of freedom and includes a three-axis gyroscope, a three-axis accelerometer and a three-axis magnetometer. Movement data were collected and recorded during the walking, running, climbing stairs and sitting movements of the volunteer by means of an inertial sensor worn on the right ankle of the volunteer. Firstly, these three recorded sensor data were synthesized and the kinematic orientation angles (pitch, roll, yaw) of the leg during the movement were calculated. Then, two property (energy and maximum value) matrices of these orientation angles were calculated. These calculated feature matrices are given as input to motion classification algorithms. Within the scope of the study, four motion classification algorithms were used. These are decision tree, k-nearest neighbor, support vector machine and random forest classification algorithms. The highest classification success in all lower extremity motion types was obtained with the nearest neighbor classifier, and the motion classification accuracy was 83.3%, 100%, 83.3%, and 91.6% for walking, running, sitting, and climbing stairs, respectively.

Keywords: