Enhancing Fitness Training with Motion Assessment Using Statistical Motion Features and Support Vector Machine

Fitness training requires motion assessment. Recently, AI-driven fitness training solutions have become available. However, most current systems use deep learning models, which necessitate a substantial amount of data. Moreover, they evaluate only some poses during training and cannot analyze motions in a certain time. They also focus on full-body motion assessment and cannot provide joint-level feedback to the user. To address these issues, we propose a fitness training system that uses statistical motion features and Support Vector Machine (SVM). Our system uses a motion capture device and a large screen. The user exercises following the virtual instructor’s motion. After each repetition of motion, the system evaluates the user’s motion, and the assessment of each joint is visualized on the screen. We employ statistical joint rotational features for analyzing motion and providing joint-level feedback. We use SVM classifiers for individual joints that can be trained with a small size of datasets. Experimental results demonstrate that our SVM-based classification model effectively distinguishes correct and incorrect motions, and our joint-level feedback is useful for fitness training.