Skip to content

pereldegla/yoga_assistant

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Yoga assistant using human pose detection in machine learning

This virtual yoga instructor tells you if you are doing the right pose.

We use Body Pose Tracking with MediaPipe BlazePose for detecting human poses and classify yoga poses using sklearn SVC.

yoga assistant using human pose detection in machine learning

You can download the dataset from Kaggle through this link .

The dataset consists of 5 yoga poses.

The core steps of this project were:

  • Building the dataset using BlazePose and Pandas methods
  • Building the model using sklearn SVM in one-versus-one
  • Evaluation of the model
  • Predicting the pose for one image
  • Predicting the pose in a video and in real-time

Video demo here:

yoga.mp4

Here are the confusion matrix and classification report for our test set:

Why do that downdog and plank get theses results ? Because they show less variations in hand gestures (check the dataset). Of course, there is room for improvement.

A further step would be using OpenPose or PoseNet instead to support real-time multi-person pose estimation. As BlazePose doesn't. Another way is using YOLO for person detection, crop the persons then predict every cropped image. The latter is a bit expensive in computational power.

Anyway this would become a virtual yoga class instructor.

My initial plan was to build a complete fitness assistant. I still plan to do it and maybe this part in it.

Connect with me on LinkedIN

About

We use the Blaze pose for detecting human poses and classify yoga poses

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages