Using Wearable Sensors and a Convolutional Neural Network for Catch Detection in American Football.

Area of Science:

• Sports Science
• Wearable Technology
• Machine Learning

Background:

• Efficient training in professional sports relies on high-volume, high-quality exercises with data logging.
• Current American football training lacks wearable sensors capable of logging pass reception events (catches or drops).

Purpose of the Study:

• To develop and validate a novel wearable sensor system for accurately detecting and logging catches and drops in American football.
• To introduce a new dataset and explore the potential for autonomous training in American football.

Main Methods:

• Utilized a nine degrees of freedom motion and audio sensor platform to collect data from 759 pass reception attempts.
• Preprocessed sensor data to train a neural network for classifying catch and drop events.
• Analyzed the significance of individual sensor signals for classification accuracy.

Main Results:

• Achieved a classification accuracy of 93% in identifying catches and drops using the developed neural network.
• Determined that acceleration and magnetometer data were most crucial for classification, while audio and gyroscope data were less significant.
• Introduced a novel dataset comprising motion and audio data from American football pass attempts.

Conclusions:

• The developed wearable sensor system effectively detects catches and drops in American football with high accuracy.
• The findings highlight the potential of sensor technology and machine learning for advancing training methodologies and performance analysis in sports.
• This research paves the way for autonomous training systems in American football by providing objective data on reception events.