Automated classification of movement quality using the Microsoft Kinect V2 sensor.

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning in Healthcare

Background:

• Traditional movement quality assessment relies on subjective, time-consuming qualitative methods.
• These methods are susceptible to inter-rater variability and measurement bias.
• Marker-less motion capture systems offer potential for objective, quantitative kinematic analysis.

Purpose of the Study:

• To evaluate machine learning models for classifying Movement Competency Screen (MCS) scores.
• To utilize kinematic features derived from Kinect V2 position data for automated movement quality assessment.
• To assess the feasibility of using objective kinematic data to replace subjective movement quality ratings.

Main Methods:

• Collected kinematic data from 31 physically active males performing squats and lunges using a Kinect V2 sensor.
• Extracted and selected domain-specific kinematic variables as input features for machine learning models.
• Employed Multiclass Logistic Regression (MLR) to classify MCS scores based on joint kinematics, validated via 10-fold cross-validation.

Main Results:

• Machine learning models demonstrated strong performance in classifying movement quality.
• Model sensitivity ranged from 0.66 to 0.89, specificity from 0.58 to 0.86, and accuracy from 0.74 to 0.85.
• Kinect-based kinematic analysis proved effective in translating movement data into objective competency scores.

Conclusions:

• Automated movement quality assessment using Kinect-based kinematic data is a viable and practical approach.
• This method offers a novel, objective alternative to traditional qualitative assessments.
• The study supports the use of marker-less motion capture for efficient and reliable movement analysis in clinical and sports settings.