Catch Recognition in Automated American Football Training Using Machine Learning.

Area of Science:

• Sports Science
• Computer Vision
• Machine Learning

Background:

• Effective training for American football receivers requires precise evaluation of individual strengths and weaknesses.
• Limited human resources necessitate automated solutions for athlete performance assessment.
• Automated passing machines provide a foundation for developing computer-based systems to analyze catch attempts.

Purpose of the Study:

• To design and validate a computer-based system for automatically recording and evaluating American football receiver catch attempts.
• To determine the success or failure of a pass reception using an automated system.
• To create a dataset and employ machine learning models for objective performance analysis.

Main Methods:

• An experiment was conducted using a fully automated passing machine and an integrated audio-visual recording system.
• A dataset of 2276 catch attempts was collected, comprising audio and video sequences.
• A Convolutional Neural Network (CNN) with Long Short-Term Memory (LSTM) was used for video data classification, and a 1D CNN for audio data classification.

Main Results:

• The developed system achieved an accuracy of 92.19% in classifying successful versus unsuccessful catch attempts.
• The combination of CNN and LSTM for video analysis and 1D CNN for audio analysis proved effective.
• The experiment successfully generated a comprehensive audio-visual dataset of receiver performance.

Conclusions:

• The feasibility of an automatic classification system for catch attempts in automated training is confirmed.
• This technology offers a pathway for more targeted and individualized training for American football receivers.
• Automated evaluation systems can significantly enhance the efficiency of athlete performance assessment in sports.