This project focuses on analyzing human body movements during common exercises by capturing and processing angles of key body joints. We utilized video data to extract frame-by-frame angles of the following body parts during various exercises such as push-ups, jumping jacks, pull-ups, squats, and Russian twists. For pose estimation, MediaPipe was used to detect body landmarks, while YOLOv6 was employed for object detection to enhance accuracy.

Methodology

• Video Collection: Videos were recorded for each exercise (push-ups, jumping jacks, pull-ups, squats, Russian twists), ensuring proper form and variety in movement.
• Frame-by-Frame Analysis: Each video was processed frame by frame, and landmarks were detected using MediaPipe's Pose Estimation. We calculated the angles of key joints by using the positional data of landmarks across different frames.
• Object Detection with YOLOv6: YOLOv6 was used to identify specific objects and enhance the robustness of the pose estimation by detecting outliers or incorrect poses during exercises, thereby improving the accuracy of the analysis.

Applications
This dataset can be used for multiple applications:

• Form Correction: By comparing these angles with standard benchmarks, feedback can be provided to improve exercise form.
• Performance Tracking: Over time, users can monitor their improvement by analyzing the changes in their joint angles during exercises.
• Pose Classification: Machine learning models can be trained to classify correct vs. incorrect form, enabling the development of smart fitness assistants.
• Real-time Feedback Systems: Using pose estimation in conjunction with live video, real-time systems can be developed to guide users during workouts.

Exercises Analyzed
The following exercises were captured and analyzed for this dataset:

• Push-ups: Key focus on shoulder, elbow, and hip angles.
• Jumping Jacks: Full-body motion tracked via shoulder, elbow, hip, knee, and ankle angles.
• Pull-ups: Primarily focused on shoulder and elbow joint movements.
• Squats: Analyzed hip, knee, and ankle angles for depth and posture analysis.
• Russian Twists: Core movement tracked via shoulder and hip angles to assess rotational motion.

Potential Analysis

• Time-Series Analysis: The data can be treated as a time-series, allowing for the identification of trends in joint movement over the duration of an exercise.
• Pose Optimization: Optimization models can be used to suggest improvements in form based on angle analysis.
• Machine Learning Integration: The dataset can serve as input for machine learning algorithms to automate form correction and workout optimization.