Boxing action recognition using inertial data and deep learning

This study explores the ongoing shift toward automated sports performance analysis by employing Deep Convolutional Neural Networks (DCNNs) for the classification of movement data in combat sports, specifically boxing. Using data from wearable inertial measurement units (IMUs), this study addresses the limitations of traditional, expensive video analysis by providing an accessible, cost-effective alternative that can be seamlessly integrated into athletes' training routines. IMUs offer a practical solution for continuous performance monitoring and feedback, establishing a foundation for automated performance assessment during training sessions. In this study, the classification accuracy and recall of DCNNs are rigorously compared with alternative algorithms, demonstrating a high overall recall of 99 % and accuracy of 91 %. Focusing on pad work training as a key application area, this work advances the automation of boxing action recognition, with implications for both training optimization and real-time scoring automation in competitive boxing. These findings underscore the potential for wearable technology and advanced machine learning methods to transform athletic performance evaluation and bring engineering innovation to sports training. • Integration of Inertial Measurement Unit (IMU) sensors. • Utilization of Deep Convolutional Neural Network to classify sports data. • Efficiency in boxing movement recognition. • Automation of boxing performance analysis.