Skip to main content

pose_estimation_fitness

Last updated by: lukgvo, Last updated on: 23/06/2025

info

Document Creation: [09/05/2025]. Last Edited: [09/05/2025]. Authors: [Luk Gvozdenac].

Document Code: [PE-1]. Effective Date: [09/05/2025]. Expiry Date: [09/05/2026].

Code Location : Redback-Operations/redback-orion/tree/main/Player_Tracking/Pose_Matching_project/Pose_Estimation/pose-estimation-fitness​

Pose Estimation Fitness Project

This project focuses on pose estimation and strain analysis for fitness applications using a pre-trained Keypoint R-CNN model. The primary goal is to detect keypoints, evaluate exercise form, and visualize strain metrics.

Summary​

Libraries Used​

  • os
  • cv2 (OpenCV)
  • matplotlib.pyplot
  • numpy
  • torch
  • csv

Key Components​

  1. Pose Estimation Model: Uses a pre-trained Keypoint R-CNN model from torchvision.
  2. Strain Analysis: Calculates strain metrics for specific exercises (e.g., deadlift, bench press, squat).
  3. Visualization: Draws poses and highlights areas of high strain.
  4. CSV Handling: Saves and loads strain results to/from CSV files.
  5. Best/Worst Form Analysis: Identifies and displays the best and worst exercise forms based on strain metrics.
  6. Graphical Output: Generates strain metric graphs for visualization.

Documentation​

Functions​

load_pose_model(device='cpu')​

Loads the pre-trained Keypoint R-CNN model.

pose_estimation(image, model, device='cpu')​

Performs pose estimation on an image.

draw_pose(image, keypoints, strain_results=None, threshold=0.5)​

Draws poses on an image and highlights areas of high strain.

calculate_strain(keypoints, exercise_type)​

Calculates strain metrics based on keypoints and exercise type.

save_strain_results_to_csv(strain_results, csv_path)​

Saves strain results to a CSV file.

load_strain_results_from_csv(csv_path)​

Loads strain results from a CSV file.

evaluate_images(data_dir, model, device, exercise_type)​

Evaluates all images in a directory and identifies the best form.

display_best_and_worst_images_with_strain(data_dir, model, device, exercise_type)​

Displays the best and worst exercise forms with corresponding strain graphs.

Usage​

  1. Setup: Install dependencies and prepare the dataDeadlift directory with exercise images.
  2. Run: Execute the main script to analyze exercise form.
  3. Visualization: View poses, strain metrics, and graphs for each image.
  4. Analysis: Identify the best and worst exercise forms based on strain metrics.

Conclusion​

This project provides a comprehensive solution for pose estimation and strain analysis in fitness applications. It can be applied to sports analytics, personal training, and rehabilitation.