Skip to content

Latest commit

 

History

History
50 lines (26 loc) · 2.03 KB

README.md

File metadata and controls

50 lines (26 loc) · 2.03 KB

PostOp Pose Tracking

This project is a part of the submission for the Future of Healthcare Hackathon by Datavant

This project tracks your position and movement during exercise to provide corrective feedback.

Examples

The system can tell if you are doing certain exercises correctly by comparing the position of each joint on the pose model.

Squats

A correct squat is calculated when the vertical position of your hips is below that of your knees.

Squats

Jumping Jacks

A jumping jack is determined to be correct when the elbow's vertical position is above your head

Jumping Jacks

Other Exercises

Feedback for other exercises can be implemented on a case-by-case basis by comparing the positions of each joint.

Technical Implementation

The system was implemented using Python with the opencv-python and mediapipe libraries.

MediaPipe takes an image of a person and generates the coordinates of 33 body parts, or landmarks, which can be used to run calculations on relative positions and drawn on the image.

MediaPipe Landmark Model

Applications

We see several applications for the use of this technology:

  1. Guidance while performing routines for physical therapy. Can be used to ensure users are exercising correctly and do not injure themselves.
  2. A virtual personal trainer that can give feedback on your exercise, to help with the obesity epidemic.

Future development

With a pool of enough data, we could create trained AI models to give specific feedback on different types of workouts without having to hard-code comparisons between joint positions.

Special Thanks

To the organizers and judges of the Future of Healthcare Hackathon by Datavant.

To the authors of OpenCV Python and MediaPipe Pose