If you don’t know how to do anything, then you need to learn some basics of deep learning and computer vision; there are some very good specializations here for that.
Now, here, in your case, it’s a complex example because it involves video as your data, and you might probably need to split that into a series of continuous frames, but it is complex to achieve. Your data are the videos of movements in the gym or therapy or, better say, the frames of those videos!