Squat Quality User Assessment Through Artificial Intelligence: A User-Specific Approach

Strength training is a method of guided exercise that improves physical well-being through repeated loaded movement patterns. One such movement pattern that is core to lower extremity development is the barbell squat. The barbell squat is a movement pattern that requires experienced scrutiny to prevent severe injuries induced by high-risk squat movement patterns. However, professional resources are limited, particularly in rural areas. In this paper, we present a machine learning framework that is user-specific for Squat Quality User Assessment Through Artificial Intelligence (SQUAT-AI). The contributions of this work include the following: (i) a data processing pipeline and structure, (ii) a variational autoencoder for anomaly detection (AD-VAE) that identifies high-risk squat movement patterns (HR-SMP) with score-based feedback, and (iii) a VAE for generating a data set with HR-SMPs, eliminating the need for a participant to ever perform HR-SMPs for training a classification algorithm. This system is meant to provide a method of user assessment without a large population.