Inspiration

Everyone wants to be fit and healthy, but working out in a wrong way can be extremely dangerous. In addition to that, personal trainers are expensive and subject to availability. Therefore, we want to make real-time gym workout feedback accessible to everyone, so that all of us can be safe and healthy.

But what the hell does Poser do?!

Poser is your own personalized AI trainer app, which monitors your posture during exercises, giving you real-time spoken feedback. You just turn our app on, select an exercise and that's it: we will help you improve rep by rep with just a little bit of motivation - if you need it!

How is it built?

Poser uses pose estimation (hence the name) to first monitor your movements. Then, it measures the angles made between particular parts of your body to verify if the exercise is being correctly done. The app has been built in Android Studio using Java and Kotlin. Also, our project uses partially the code of the following TensorFlow model (mobilenet-v2): https://github.com/edvardHua/PoseEstimationForMobile

Challenges we ran into

Initially, it was very hard to find a viable pose estimation framework. We almost abandoned the idea before coming across a mobilenet-v2 model integrated well with Android. After the long process of setting up the pose estimation we found ourselves in a slight pickle. It was very difficult to rigorously and mathematically describe the indicators which determine if the exercise is being done properly. Camera-based apps are also notoriously hard to design. They always end up looking unnatural and ungainly. Thankfully, we were able to seek help of Sam from American Express, who expertly guided us through the visual aspect of the app, making the user experience top-notch.

Accomplishments that we're proud of

This is an app we ourselves would honestly love to use, which is exactly why it is wonderful to get a working proof-of-concept. Not only does the app perfectly guide you through the exercises, it also helps you forget the tedious task of counting reps.

What we learned

We definitely learned about our physical limits and capacities.

What's next for Poser

Hopefully, we can integrate even more exercises into the app, while using more sophisticated and automated methods of monitoring the workout. This would entail having a target skeleton and a well defined distance function.

Built With

Share this project:

Updates