sports_pose_scripts

Pose estimation with motionBERT

This is a pet project for 3D pose recovery for table tennis teaching.

It compare student's pose with a teacher's pose, and output a score base on similarity in positino and speed.

Core pose recovery processes come from motionBERT and Alphapose.

Demo

Input video: https://www.youtube.com/watch?v=Ex5E1iLzsSw

Output:

(Hit moment is automatically detected based on speed of hand motion.)

Installation

Conda Environment

1. Install Anaconda

2. Create conda environment using the provided environment file:

conda env create -f environment.yml

3. Get motionBERT and Alphapose from their repositories.

Folder Structure

Follow the directory hierarchy below for organizing your project files:

- sports_pose_scripts
- motionBERT
-- ...
- AlphaPose
-- ...
- projects
-- [project_name]
--- dataset
---- table_tennis.mp4

Running the project

Navigate to the sports_pose_scripts folder in your terminal

Run the following command to execute the project:

python run.py project_name