Football teaching and training based on video surveillance using deep learning

Area of Science:

• Sports Science
• Computer Vision
• Machine Learning

Background:

• Objective athlete performance evaluation is crucial for elite sports research.
• Automatic identification and classification of football exercises using video monitoring overcome manual analysis limitations.
• Video analysis aids in detecting and preventing inappropriate actions by classifying digital video material based on human actions.

Purpose of the Study:

• To systematically apply data from Inertial Measurement Units (IMU) and computer vision for Deep Learning of Football Teaching Motion Recognition (DL-FTMR).
• To analyze training through deep learning models and assess the efficiency of coaches for sport-specific decision-making.
• To develop and validate a DL-FTMR system for enhanced football training analysis.

Main Methods:

• Utilized data from IMU and computer vision analysis for DL-FTMR.
• Explored various libraries and deep learning methods for profound model construction.
• Employed video-based research and the UT-Interaction dataset for complex event identification and Human Activity Recognition (HAR).

Main Results:

• DL-FTMR achieved a 94.5% performance ratio, outperforming OCNN, GMM, YOLO, and HAR-SAM.
• High ratios were reported for behavior processing (92.4%), athlete energy (92.5%), interaction (91.8%), prediction (92.5%), sensitivity (93.7%), and precision (94.86%).
• Video monitoring systems provide a game-like view, enhancing selection accuracy perception without compromising response time.

Conclusions:

• Practicing with video monitoring systems enhances selection accuracy perception.
• DL-FTMR offers a valuable technique for objective athlete performance evaluation in elite sports.
• The study validates the effectiveness of integrated IMU and computer vision data for advanced motion recognition in football training.