Radar-based Machine Learning Approaches for Classification of Rehabilitation Exercises

Abstract

Muscular rehabilitation is essential for injury or surgery recovery by restoring strength, flexibility, and range of motion to the affected joints and muscles. It can also improve posture correction and performance by strengthening weak areas, reducing the risk of injury, and managing chronic conditions like arthritis, osteoporosis, or chronic pain. Currently, there is only physical therapy for these problems, and the treatment is in person at a specific location, such as a hospital or a clinic. Other works proposed mounting surface electromyography to recognize muscle activation patterns or wrist-forearm for muscle fatigue or using cameras to video call for sessions. Regrettably, such works put the patient’s privacy or comfort in danger. Our proposed solution is a radar and machine learning-based monitoring and classification of rehabilitation exercises. This RF-based system can accurately monitor and classify exercises that are part of the treatment for a specific need and in the privacy of the patient’s house. The proposed solution uses the RF reflections of the body and the environment. It uses these signals to analyze them in a machine learning algorithm to classify the exercise the person executes. This solution could be used anywhere in the home by any patient with minimal setup effort. Our results, done by four subjects in their own homes, show that the already trained model can classify with an accuracy of 87% to 97%.

Date of Award	Jun 2023
Original language	English (US)
Awarding Institution	Computer, Electrical and Mathematical Sciences and Engineering
Supervisor	Mohamed-Slim Alouini (Supervisor)