Inspiration
While working out recently in my gym and doing weighted squats I injured myself due to bad posture which I didn't realize. As a result of that, I had to deal with pain and physiotherapy for the next 2 months. I had a physical trainer (which is pretty expensive) who had taught me what to do, but In my day to day practice missed some critical steps and ended up injured. The requirement is continuous monitoring, coaching, and motivation.
What it does
Using modern deep learning techniques the app is able to interpret the live video stream and locate the position/coordinates of various body parts which is fed into a rules based engine which 1) Keep count of the exercise being performed 2) Matches the realtime posture with the rules for that particular exercise to check if everything is being done correctly.
Challenges I ran into
Training the model and saving it locally for real time inference on the mobile device
Accomplishments that I'm proud of
Using transfer learning to augment the POSENET model from tensorflow to be able to understand various exercises and detect anomalies when it deviates from the defined rules
What I learned
Training deep learning models on powerful severs and using them locally for inference on a low compute capable device for real time work.
What's next for Gym assist
Built With
- javascript
- python
- react
- tensorflow
Log in or sign up for Devpost to join the conversation.