Real-time AI-assisted visual exercise pose correctness during rehabilitation training for musculoskeletal disorder

Exercise therapy is a prominent method for recovering from musculoskeletal disorders like Work-related Musculoskeletal Disorder (WMSD), Classroom-related Musculoskeletal Disorder (CMSD). This method requires the monitoring of a physiotherapist to advise the performer to do the exercise poses correctly. This process is costly and requires the presence of a physiotherapist. In this work, we plan to develop software for the real-time detection of exercise poses with the help of AI. This work replaces the physiotherapist and provides good real-time feedback on the users' wrist extension and flexion exercise posture. Artificial Intelligence (AI) is the most popular and widely used technique for real-time image analysis. In real-time the user takes at least 1 s to perform the exercise pose. Our model classifies and provides feedback within 0.79 s. The frame processing rate of our model is ~ 21 frames per second. Experimentation of this framework was done through CNN DenseNet with 2-level architecture. We have experimented with three different ways of outcomes. Our mode achieved 100% accuracy with 106 samples of data and 99.86% accuracy with 2160 samples. Finally, with 144 cross-person sample datasets achieved 83.33% accuracy. This technique performs well for evaluating wrist extension exercise poses for recovery and gives participants immediate feedback on whether their wrist extension is correct or incorrect.