Form-ulation

Inspiration

Our friend recently started going to the gym. When he's with us, we can tell him the correct form. When he goes alone, he doesn't know if he's using the right form, so we decided to build an AI to help him.

What it does

Our project is a Real-Time AI Fitness Coach specifically designed to analyze your exercise form using Computer Vision. Instead of just recording a video, the program "sees" your body as a collection of data points and uses geometry to check your form

How we built it

We are using YOLOv8, a pre-trained CNN, to detect joints and ligaments on a person's body. Using this, we create our own dataset with points and angles mapped out for good and bad form by writing a simple Python script and analyzing videos. We use our dataset to train an MLP (Multilayer Perceptron) to distinguish between good and bad forms.

Challenges we ran into

The hardest problem we faced was finding and collecting data for our dataset The Python version was incompatible with our original library, so we had to find a new computer vision library.

Accomplishments that we're proud of

Built our own dataset Used computer vision map out the points on the human body We were able to map out the position of joints on the image

What we learned

We learned how to create our own dataset and apply computer vision. We also learned how to train our own AI model

What's next for Form-ulation

We currently have a limited number of exercises in our dataset. After this hackathon, we want to expand our dataset so that it can check your form for every exercise