Template matching based motion classification for unsupervised post-stroke rehabilitation

Post-stroke rehabilitation training is proven clinically to be essential and effective on helping stroke patients to regain part of the functionality of their body. In recent years, cost-efficient rehabilitation training programs especially unsupervised training have become a popular research area due the increasing number of post-stroke hospitalizations and the high healthcare expenditure associated. In order to achieve unsupervised rehabilitation training, a reliable continuous monitoring measure is crucial. This paper proposed a motion classification system based on template matching technique that can constantly identify and record the quantity and quality of patient's rehabilitation exercise as a reference for the professionals to analyze patient's recovery progress. It can also integrate features like fall detection to improve safety in unsupervised training environment. In contrast to the conventional motion tracking system which are generally expensive and complicated to operate, the proposed system uses only low-cost non-visual based wireless sensors for acceleration data collection. Since the classification process is based on template matching, there are no additional sensors like gyroscope required for precise reconstruction of patient's motion. To test the system performance, a preliminary experiment involving an actual stroke patient has been conducted. Despite the movement performed by the patient was non-standard and inconsistent, the system was still able to identify the predefined exercises from a series of movements and count the number of repetition for each exercise accurately.