This is a pretty fertile area of research between ECE and BME. I'm reasonably sure that most of the research professors doing this kind of thing are ECE, but ECE and BME are pretty close as departments, and there's a lot of joint papers being submitted to IEEE EMBS on the subject. Of course, there may be 3 different CoC labs doing the same thing and no one in the CoE knows about it, because... institutional politics.
A lot of what the article talks about is 10 year old tech that anyone with 2000 level classes in programming and signal processing could recreate. There's even an open source library for that.
The problem is that there's so much noise introduced by the detection method and so little data being collected that you don't really wind up with a uniquely useful tool. Can it show you bad jumping form? Absolutely. Can it show you bad form that a physical trainer watching the same video couldn't identify? Probably not, at least not right now. Some of the more intriguing research involves using algorithms that track many, many variables and iteratively choose the maximum explanatory variable, minimum covariance models. That may well push this technology a big step further.
However, keep in mind that this is a tool for analysis, not a scrying stone for perfect throws. Finite element models many, many time more accurate than the basic 'hips, knees, ankles, feet' are used to iteratively improve aircraft models, etc. But it's small and diminishing improvements- you still have to do design work. And these are people, not airframes. Strengthening a muscle is straightforward. Relearning a motion less so.
The really exciting area (shameless plug for my own research here) is in networks of sensors. The human eye can get a good 95% accuracy on slow motion video. The remaining 5% that quantifying that video gets you might not be all that important by itself, but if you combine it with a pulse oximeter and ballistocardiogram or ekg and bioimpedance data you can start to find some pretty interesting things that is really not obvious from each data set individually. My research is in disease detection, but I can totally see quantifying biosignals in a game environment as a huge advantage. We all think we're heroes who will win the game, but what if there was a data set on our right guard that shows his performance changes in the following concrete ways as his pulse ox drops. Makes substituting him a lot more clear a decision, right? Even more so, a lot of injury comes from bad form caused by degraded performance, so you'd be able to see red flags on guys who are so tired they're playing dangerously.