Inhaltszusammenfassung

The role of computer-aided therapy in physical rehabilitation has significantly grown in recent years. In order to optimize the therapeutic actions based on the patient disease and improve the interaction between the patient and telemedical software, it is important to differentiate the healthy individuals and patients following different chronic diseases or musculoskeletal disorders. In this paper, we propose a deep learning method to classify the trajectory patterns in physical rehabilitati...The role of computer-aided therapy in physical rehabilitation has significantly grown in recent years. In order to optimize the therapeutic actions based on the patient disease and improve the interaction between the patient and telemedical software, it is important to differentiate the healthy individuals and patients following different chronic diseases or musculoskeletal disorders. In this paper, we propose a deep learning method to classify the trajectory patterns in physical rehabilitation as healthy or pathological. Motion sequences containing joint positions and joint angles are transformed into an image representation, which enables the training of a classification model using a deep 2D Convolutional Neural Network (CNN) to infer a health state or a symptom. Our approach was evaluated on two publicly available datasets, the KIMORE [1] dataset for complete body skeleton sequences and the Toronto Rehab Stroke Posture (TRSP) [2] dataset containing upper body skeleton sequences. The current method effectively classifies the healthy and pathological movements and achieves 78.57% of accuracy on KIMORE dataset and 86.20% of accuracy on TRSP data.» weiterlesen» einklappen

Autoren