Motion Recognition for Unsupervised Hand Rehabilitation Using Support Vector Machine

In recent years, with the rapid increase in stroke patients and the associated cost, efficient stroke rehabilitation especially unsupervised and remote stroke rehabilitation have become hot research topics. It has been proved that unsupervised stroke rehabilitation was effective and necessary for stroke patients. However, an accurate and robust classification system for hand motion recognition is essential for such an unsupervised system. In this paper, we present a support-vector-machine-based finger and wrist movement recognition system designed to identify typical hand training movements such as finger docking, cylinder grabbing and sphere grabbing. Three stroke patients were involved in this clinical research. For each training movement, 35 different movements from those three patients were recorded respectively to verify and validate this system. The data were separated into two groups; one training and one testing group. After preprocessing and feature extraction of the acquired motion data, the support vector machine recognition approach was employed to establish a small sample identification model. Finally, the data of testing group were used to verify the developed model. It was found that the recognition accuracy of the developed model was 96.67. This research paves the way for development of an automated system for stroke patient rehabilitation.