Towards the Development of Human Action Recognition and Monitoring System for Rehabilitation Purposes: A Feasibility Study

According to the World Health Organization, the growing number of individuals affected by neurocognitive diseases (e.g. Alzheimer, Dementia, Parkinson) is expected to rise and there will be an increased risk of hospitalization and an increase necessity to do rehabilitation at home as well as to stay active. The primary aim of this paper is to develop a Human Action Recognition (HAR) system that can be used at home by the patient as both a portable fitness tool and personal coach. The system’s primary function is to suggest personalized exercises and supervise the proper execution of movements. Additionally, clinicians can monitor patients’ progress on a daily basis through a set of biomechanical key performance indicators (Bio-KPIs). The ultimate goal of the system is to empower patients to improve their physical health and allow clinicians to track their progress over an extended period of time. We conducted a test of the network using 10 participants who performed 4 rehabilitation exercises, resulting in a test accuracy of 78.45%, a precision of 87.75%, a recall of 84.7%, and f1-score of 82.5%. The performance is quite encouraging since we tested the system in real environments under different light conditions and we obtained performance values that are comparable with the literature. Moreover, the Bio-KPIs extraction let us characterize the rehabilitation exercises.