AI-assisted Automatic Jump Detection and Height Estimation in Volleyball Using a Waist-worn IMU.

Area of Science:

• Sports Science
• Biomechanical Engineering
• Machine Learning Applications

Background:

• Volleyball jump analysis is crucial for injury prevention but manual methods are labor-intensive.
• Inertial Measurement Units (IMUs) and machine learning offer efficient alternatives for jump analysis.
• Existing research often separates jump classification and physical load estimation, lacking integrated solutions.

Purpose of the Study:

• To develop an automated pipeline for detecting volleyball jumps and predicting jump heights using waist-worn IMU data.
• To integrate jump detection, classification, and height estimation into a single system.
• To provide a practical tool for monitoring player load and reducing injury risk.

Main Methods:

• Utilized a Multi-Stage Temporal Convolutional Network (MS-TCN) for jump segmentation and classification from time-series IMU data.
• Employed three downstream regression machine learning models for jump height estimation based on detected jump segments.
• Validated the pipeline on a dataset of 10 players and 337 jumps.

Main Results:

• The system achieved high accuracy in identifying jump activities and their types (F1-score = 0.90).
• Demonstrated superior jump height prediction performance (R-squared = 0.53) compared to the commercial VERT device (R-squared = -1.53).
• Successfully integrated jump detection and height prediction into a cohesive pipeline.

Conclusions:

• The developed pipeline offers an accurate and efficient method for analyzing volleyball jumps using IMUs.
• This integrated solution provides a valuable tool for monitoring physical load and mitigating injury risks in athletes.
• The findings highlight the potential of machine learning and IMUs for performance analysis and injury prevention in sports.