Automatic recognition of physical exercises performed by stroke survivors to improve remote rehabilitation

Strokes are the second cause of death and the third cause of disability in the world. Currently, there is not an actual cure for stroke victims, but physiotherapy can be used to restore as much mobility as possible until a plateau is reached again. However, performing these exercises implies that both patient and therapist are together in the same place, so that the latter can guide the former through a correct execution of physical exercises. This raises additional difficulties when it comes to continuously monitoring the recovery of stroke patients, due to the economic costs involved and the requirements of both geographical and temporal availability. These issues can be addressed by leveraging technology, specifically computer vision-based assistive systems and remote rehabilitation tools, so as to the affected person can check whether the exercises are being performed correctly. This paper is focused on the automatic classification of exercises, within the context of a gamification-based remote rehabilitation tool used to automatically assess the performance of stroke patients when making physical rehabilitation. To this end, we use the DTW algorithm for analyzing and comparing open-ended motion curves, so that the exercises do not have to be fully performed until existing candidates can be matched. This increases the system flexibility and offers an interaction mechanism much more simple, which is usually a needed requirement by stroke patients.