FORMe

Inspiration

As beginners ourselves, we faced the problem of not wanting to go the gym due to time constraints, expenses, and shyness, but still needed something to guide us through our workouts and correct us if our form was wrong.

What it does

It gives you personalized workout plans and checks your form as you work out, giving live feedback to help you fix your form.

How we built it

We used MediaPipe's blazepose for the skeletal overlay, and then ran a Random Forest model on data we found online. To add more features, we turned to a geometric math based model that let us set parameters and checked form accordingly.

Challenges we ran into

The correct angle for the camera to see all parts of the body was very challenging to find, and we often got false positives as the camera would incorrectly estimate parts of our body. The pi also did not have enough RAM to run our model, so we had to switch to a lite version, and run some software on the screen device. We also tried to implement a servo motor to move the camera, but this did not work due to hardware issues.

Accomplishments that we're proud of

Running two different models to compare accuracy and better fit. Including a variety of different exercises, and overall having fun and learning a lot. We also redefined our product after our first round of market research, leading to a product we were all a lot happier with.

What we learned

None of us had a lot of coding experience, so we all learnt a lot in terms of the technical aspects of this project, and we also learnt about actually listening to your market to define your product.

What's next for FORMe

We want to add more hardware aspects such as a bracelet with IMU sensors to track your motion and provide more data. We also want to expand to other fields such as dance and martial arts.

Built With

• autodesk-fusion-360
• claude
• mediapipe
• vscode