As I said, the easiest way would be to use reflective sensors and capture the image with a camera, then process the points where the reflectors are attached. You will need to grab a frame, scan for these points, and process it. This will not work for anything the camera cannot see.
I suggest this method because of its simplicity, and because of it being cost effective for people who have webcams and etc. There would be some compatibility issues, but I am not sure about that sort of thing.
You can use tri-axis accelerometers, which can detect fairly accurately motion in all directions. This could accurately model movements...
You need to ask yourself, do the ends justify the means? This sort of thing is a major headache, but if you feel that there is a lot of worth doing this, then I guess it will give you enough drive to invest your time and money into it.
Steve