Classification of squat quality with inertial measurement units in the single leg squat mobility test.

Many assessment and diagnosis protocols in rehabilitation, orthopedic surgery and sports medicine rely on mobility tests like the Single Leg Squat (SLS). In this study, a set of three Inertial Measurement Units (IMUs) were used to estimate the joint pose during SLS and to classify the SLS as poor, moderate or good. An Extended Kalman Filter pose estimation method was used to estimate kinematic joint variables, and time domain features were generated based on these variables. The most important features were then selected and used to train Support Vector Machine (SVM), Linear Multinomial Logistic Regression, and Decision Tree classifiers. The results of feature selection highlight the importance of the ankle internal rotation (IR) angle in classifying SLS. Classification results on a human motion dataset achieved an accuracy of 98% for the two-class problem using SVM, while for 3 class classification, the maximum accuracy was 73% using Decision Tree.