Human pose estimation in physiotherapy fitness exercise correction using novel transfer learning approach

To introduce and evaluate an efficient neural network approach for human pose estimation and correction during physical therapy exercises using wearable sensor data.

We leveraged benchmark data consisting of 276,625 records from wearable inertial and magnetic sensors. A novel method termed Random Forest Long Short-Term Memory (RFL), which integrates long short-term memory and Random Forest neural networks, was implemented for transfer feature engineering. The smartphone sensor data was used to generate new temporal and probabilistic features. These features were then utilized in machine learning methods to classify physical therapy exercises. Rigorous experiments, including k-fold validation and hyperparameter optimization, were conducted to validate the performance of the RFL approach.

The RFL approach demonstrated superior performance, achieving a remarkable 99% accuracy with the Random Forest method. The rigorous experiments confirmed the efficacy and reliability of the method in classifying physical therapy exercises.

The proposed RFL method introduces a novel feature generation approach enhancing the accuracy of physical therapy exercise classification and correction. This innovative integration not only improves rehabilitation monitoring but also paves the way for more adaptive and intelligent physiotherapy assistance systems. By leveraging sensor data and advanced machine learning techniques, it has the potential to mitigate risks associated with disabilities and major diseases, thereby offering a feasible alternative to frequent clinic visits for consistent therapist guidance.