I am building a database for human movement. Right now each exercise or pose or movement is indexed by its name - Downward dog, squat, handstand, etc. But this gets confusing when multiple names apply to the same movement. The true identifier is the motion of your limbs in space. I want to encode that motion/position so that if you try and upload L-sit handstand and also L-sit, it tells you that you are try to upload a duplicate (except for the arms). Furthermore, hopefully each movement can be uniquely encoded into an ID that could be used at a webURL. If you did this you could also compute the similarity between two movements to indicate that one is a progression step to train the other.
I don't have a computer science background so encoding an compression are new to me, but Im a good hacker and I can quickly get things like open pose working. Im trying to complete this in the next 3 months. Wish me luck.
Yes, exactly. Thanks for reminding me of that - I remember reading about it a few months ago. Hopefully I could encode a user uploaded video into Laban Movement notation basically to index it in the database.
Yes, I am considering using exrx for the base training set actually. They have an api where you can get movement labeled (which muscles are contracting), license video footage of many exercises. Maybe I can somehow integrate what Im building with their site.
I like the idea, but I'm curious what sort of things you are hoping to accomplish.
For instance is human deformity important? Do you want to take into account mild to severe birth defects or dismemberment? How about things like severe burn scars or other medical conditions which can limit maneuverability?
Limitation of movement is a huge aspect of what I want to do with it. Everyones body is different. Thus our hips can rotate different amounts in different orientations. Thus if you submit footage of yourself doing a yoga flow for instance, the database could understand the end ranges of motion - seeing that through all motion, the angle never exceeded 38 degrees for instance.
Then a year later you could submit new footage of a flow. The software would upload all movements into the database and recognize that that angle had increased. You now have increased movement.
Now lets say someone is missing an arm for instance. When their body is in a pike position, the database should still recognize the pike and index it against other pikes (both that user and the global range of pikes). So the database is going for similarity, not exactness. Relating similar movements though making them distinct is what I believe will be one of the hardest parts of this challenge.
I believe a large part of health and fitness is understanding our own bodies and the movements we can perform. Whether its a total of 25 possible movements or 500 possible movements, exploring and training those movements is always cathartic.
I am building this for a yoga/bodyweight training workout tracking app that I have made, but was having trouble indexing all of the motions for. Though I think if it works properly, the applications could be greater than the app.
Is there a way to get updates with your project's progress?
Human movement notation is also a hard problem for dance and choreography that needs solving. My partner has been talking about it for years, and there would definitely be interest from the dance world.
I don't have a GitHub repo or anything going yet. I can update this comment when I do. My email is in my profile if you want to stay in touch. In fact, I would really appreciate it. Dance and choreography would certainly be a big aspect.
I was not. thanks for the link. Often I've found these datasets (this reminds me of CMU's motion capture dataset for instance) are more for everyday movement - things like the sitting down in a chair or calling for a taxi. Though their charter seems to imply looking for all possible human movement.
I don't have a computer science background so encoding an compression are new to me, but Im a good hacker and I can quickly get things like open pose working. Im trying to complete this in the next 3 months. Wish me luck.