LOWER LIMBS MOTION MONITORING IN THE EXECUTION OF REHABILITATION EXERCISES USING INERTIAL SENSORS AND MACHINE LEARNING

Abstract

This research project explores the integration of Inertial Measurement Unit (IMU) sensors and Machine Learning (ML) algorithms for analyzing human motion during physical therapy exercises. The monitoring of rehabilitation is crucial for assessing patient progress accurately and objectively, allowing for adjustments to treatment plans as needed. Using sensors provides continuous and detailed data on patient movement during daily activities, offering a more comprehensive insight into their health status compared to periodic assessments. Experimentation focuses on comparing predictions from IMU data with ground truth obtained from the optical system. Results show MLP and CNN-LSTM models outperforming others, with a positive effect of increasing window size on model performance. When evaluating the generalizability of models to new volunteers, MLP performs most effectively in terms of RMSE and MAE. Overall, the results support the effectiveness of MLP and CNN-LSTM models in predicting Lshin segment rotation angles accurately, indicating the importance of window size on model performance. These findings align with existing literature, emphasizing the potential of IMUs and ML models in improving rehabilitation techniques and patient care.