It all began when I started seeing kids at the gym trying to get swole (i.e. extremely muscular) without understanding the fundamentals of basic exercises. Etched into my memory is the case of little Timmy who was doing bicep curls so bad I had to intervene to prevent him from getting injured. From that day forth, I vowed to do something so that other kids like Timmy would never have to risk serious injury just to get in shape. Well folks, this weekend we did that "something." It's called SwoleSafe.
What it does
It lets the user to create a profile that describes the user's physical attributes as well as workout interests. Based on personal information, a list of applicable exercises will be generated through an expert-crafted recommendation system and rendered on the training page. The training app will detect user's motion and check to see if the users are performing the exercises in the correct form and warnings will be shown if done incorrectly.
How we built it
The technology was built using state-of-the-art deep learning for pose estimation and motion tracking. We created a REST API to allow the client application to add users/goals to a SQL database and retrieve custom dynamically-generated exercise routines for each user.
Challenges we ran into
Not very familiar with React. Got a lot of trouble with connecting pages and integrating machine learning libraries.
Accomplishments that we're proud of
We made the pages look very nice. The detector also works pretty well and real-time too.
What we learned
We learned how to use Google Cloud VMs, Tensorflow.js, React, Google Cloud Recommendation Engine.
What's next for SwoleSafe
This is just the beginning for SwoleSafe, folks. Obesity is one of the great challenges of our time. SwoleSafe will allow millions of users to get in shape without injuring themselves and help athletes fine-tune their training to get maximum gains.
In terms of the technology, we plan on collecting information about user workouts to fuel a more advanced recommendation algorithm.