One of the biggest challenges when it comes to aging in place is the lack of physical activities! Exercise in Place is inspired by the idea that seniors can be independent and exercise with the supplemental help of a computer.

What it does

Exercise in Place is a website that contains stretches and movements to exercise in place. After selecting a stretch, the user is presented with an image to mimic and their webcam recording highlighting the posture. Throughout the time of the stretch, Exercise in Place constantly checks if the posture is valid for that exercise to minimize the risks of injuries through the activity. Not only that but, Exercise in Place gives instant feedback to the user about his current posture and progress. By doing that, Exercise in Place is able to make exercising in place for seniors easier, safer, and more effective!

How I built it

The front-end of the project was built with HTML, CSS, and Javascript. The backend was built on a Python Flask server and the whole application is deployed on Google Cloud. The Machine Learning model was fed with our own dataset on a Google Cloud Bucket with the pictures, poses, and exercises. The model was built with Tensorflow in Google's Pose teachable machine. The evaluation of the video-stream happens in the Javascript with the Tensorflow library.

Challenges I ran into

Since Exercise in Place wanted to detect the exercises, find problems, detect posture, and avoid potential injuries, a custom dataset was needed to make sure the model worked. This took some time to make due to the adjustments of the dataset, WebGL crashes while building the model, deployment errors, and bad upload velocity.

Accomplishments that I'm proud of

I'm proud of developing a really cool platform that uses Machine Learning to facilitate exercises for people that are aging in place!! It was amazing to see the pose algorithm functioning correctly. I'm also proud that I developed my first dataset with images.

What I learned

I learned to be patient and to restart my computer multiple times (before the crashes). Besides that, I learned how to create a Machine Learning model through Google's Teachable Machine and export it using Javascript. I learned how to extract frames from a video to build a dataset.

What's next for Exercise in Place

Exercise in Place has a bright future on its path. The next step is to make a collaboration tool that lets professionals upload valid and invalid exercises to improve accuracy and extend benefits. Submission:

Share this project: