Sportify

User interface - upload image to analyse
Processing an image and adding 'landmarks'
Pose analysis output - ready for training

Inspiration

Our inspiration comes from the proliferation of people exercising at home, or who are in need of exercising at home because of the recent COVID 19 pandemic. And with restrictions gradually lifting, many are getting back into the habit of exercising, both at home or in a gym. Form is important because it can make the exercise easier, more efficient and give results, whereas bad form can lead to potential injuries and long term physical damage.

It is difficult to verify your form without your friend or a personal trainer, which not everyone can afford, so there is usefulness in an application that can analyze your form at any moment anywhere.

What it does

Sportify is a web application that uses computer vision to track users joints during a workout, and then compares them to a model trained on Olympic athletes to identify if something is wrong with your form. If form is poor, then landmarks will be red, and if its good, landmarks will be blue.

How we built it

Flask, Python, HTML5, CSS for the front-end, Javascript, Pandas, OpenCV, MediaPipe, Scikit-learn for backend

Challenges we ran into

No well-formatted dataset available online, we had to improvise and hand-pick our own training data. Led to immature model One of our team members dipped last-minute, so we were down a person.

Accomplishments that we're proud of

Proud of the front-end and logo, I think the visual design turned out nice. Built my first ML pipeline

What we learned

Learned flask, Javascript, improved my web design skills, with things like implementing buttons to attach a file.

What's next for Sportify

Since Sportify only works on weightlifting, we can improve the model by expanding it to cover more, including body-weight exercises which can be done by anybody at home. This would involve gathering good quality data about good and poor forms for the aforementioned exercises.