Application of recurrent neural networks for motion analysis with OpenSim

• Main Author: Yam, Li Hao
• Other Authors: Yifan Wang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Physics; OpenSim; Gait; Motion analysis; Recurrent neural networks; Machine learning; Data science; Long short-term memory networks
• Online Access: https://hdl.handle.net/10356/177883
• Institution: Nanyang Technological University

This paper explores the application of Recurrent Neural Networks (RNN) – Long Short-Term Memory (LSTM) networks in particular, to predict muscle force and movement during stoop lift motions with data processed using OpenSim software. The study includes machine learning techniques such as biomechanical data analysis, feature selection via Random Forest and hyperparameter optimization with Optuna study to improve the accuracy of the LSTM model. The created LSTM model is able to process complex sequential biomechanical data which has the potential to positively impact stroke patients’ lower limb rehabilitation. By providing more precise and individualized treatment insights, machine learning approaches can prospectively revolutionize rehabilitation practices in the 21st century.