A Framework for Assessing Physical Rehabilitation Exercises

Physical rehabilitation exercises effectively reduce the utilization of healthcare systems through exercises designed to restore and improve the level of functionality of each patient and track the recovery process. Thus, the rehabilitation exercises lower hospital admissions, length of stays, and readmissions. This study proposes a framework for evaluating physical rehabilitation exercises and monitoring patients' progress to reduce the costs associated with rehabilitation. The automated evaluation of the exercises also provides personalization options that enable clinicians to design personalized treatment plans. A fully-automated evaluation framework must identify the distinct skeletal parts, angles, and trajectories for each exercise to distinguish one exercise from another. The proposed framework starts by recording a patient's movements using an RGB camera and then extracts the different skeletal parts from the video for classification and monitoring purposes. Thus, the study also addresses the feasibility of employing an RGB rather than a Kinect camera. A feature ranking algorithm (the Fast Correlation Based Filter) was applied to select essential features. Then, the experiments investigated two classifiers using two sets of rehabilitation exercises (along with their respective mistakes) to classify the movements. The proposed approach achieved 99.64% and 90% accuracy using the Extra Tree Classifier and the One Dollar algorithm, respectively.