Skip to content

Posture Monitoring build on CNN & ANN using TensorflowJS and Movenet3 with client-side rendering

License

Notifications You must be signed in to change notification settings

Garvit9000c/PosMate

Repository files navigation

POSMATE

"Have a Buddy That Guides you for the Best, & Manages Your Posture Better"

Click Here for Presentation

Click Here for Video

Problem Statement / Inspiration

  • "Are you sure you are sitting in a correct posture right now when you are looking at your desktop screen reading this problem statement?" No, right?

  • "With the underlying Pandemic at place, Programmers/Students/working professionals are spending more Time with there laptop, working out day and night!"

"The wrong Posture of Sitting with Minimal Breaks In this Digital Era with Average Screen-time More than 5 hours, are deteriorating Life quality Standards by leading to problems related to Spinal cord, gastrology, weakening of eyesight and many more consequences which eventually affects the work-life which we were onto”

Our Hack / What it does!

A Robust AI-Driven System (Web App) That Monitors :

  • Posture (7 Body KeyPoints)
  • Relative Distance from Display

Which Notifies User when his/her Posture or Relative Distance is Wrong with a Suggestion on how to improve it! Using

  • CNN And ANN Model
  • Tensorflow.js/ Movenet3.0 with Client Side Rendering.

Algorithm / How we Build it!

Challenges We Ran Into!

Detecting Body Coordinate with Computer-Vision Efficiently.

  • 7 Body Keypoints Detection with Movenet3.0 CNN Model with 30-40fps frame rates.

Classification of Each Sample with More Accuracy in different Environments.

  • Artificial neural networks for Posture classification

Effective Alerts about Wrong Posture.

  • Smart Sound Alerts with text Notifications.

Making It Easier For Diverse User, to Use Our App!

  • Minimalist UI/UX

Server-Load & Data Breach/Consumption Problem.

  • Client-Side AI rendering which reduces server load Drastically with no Data Going Over Server.

What we Learned:

  • Deploying AI model at Client-Side with Tensorflow.js
  • Artificial Neural Network Classification & Convolutional Neural Network
  • Automated Data Collection & Labeling with Google Forms

What Next:

  • Scaling the Web-app to Chrome Extension.
  • Automated Exercise Guidance Model.
  • Monitoring Screen Time for Better Suggestions.

Technologies Used / Build With:

Python/Flask/Heroku

HTML5/CSS3/JavaScript/Bootstrap

Tensorflow.js / SkLearn / Movenet

Try it Out

PoseMate Deployment