The aim of the research was to conduct experimental studies on the recognition of single punches in boxing based on the MediaPipe framework, which includes models of convolutional neural networks BlazePose to determine a person's posture.
Methods and organization of the research. To recognize movements and the type of a punch, the open source MediaPipe framework was used, which has a built-in convolutional neural network BlazePose to determine the movements and a posture of a person. The studies were conducted with the participation of 14 athletes with training experience of 2-3 years. There were 14 men in the control group, age – 21 ± 3 years old, weight – 70 ± 12 kg, height – 175 ± 11 cm. Only straight punches that were applied to the right and left with a change of positions were investigated. The recognition of punches was carried out on the basis of the k-nearest neighbors algorithm (k-NN).
The research results. Studies have shown that the accuracy of recognition by the left hand (0.86) and the right hand (0.87) is lower than the same indicators for cross–counters – 0.95 and 0.96. The average accuracy for all classes was 0.93. The accuracy of the classification results obtained was evaluated on the basis of the F-score.
Conclusion. The simplicity and effectiveness of the technique allow implementing it into coaching practice, and make it promising for further research of other types of punches.
- Bazarevsky V., Kartynnik Y., Vakunov A., Raveendran K., Grundmann M. BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs. doi = 10.48550/ARXIV.1907.05047.
- Bhargava, Neha & Cuzzolin, Fabio. (2020). Challenges and Opportunities for Computer Vision in Real-life Soccer Analytics. Conference: AAAI 2020 - Workshop on AI in Team Sports (AITS 2020)At: New York, USA.
- Campos, Guilherme O.; Zimek, Arthur; Sander, Jörg; Campello, Ricardo J. G. B.; Micenková, Barbora; Schubert, Erich; Assent, Ira; Houle, Michael E. (2016). «On the evaluation of unsupervised outlier detection: measures, datasets, and an empirical study». Data Mining and Knowledge Discovery. 30 (4): 891–927. doi:10.1007/s10618-015-0444-8. ISSN 1384-5810. S2CID 1952214
- D. Dinu, B. Millot, J. Slawinski, J. Louis. An Examination of the Biomechanics of the Cross, Hook and Uppercut between Two Elite Boxing Groups. Proceedings, 2020; Volume 49, pp. 61. https://doi.org/10.3390/proceedings2020049061.
- D. K. Kimm, D. Thiel. Hand Speed Measurements in Boxing. Procedia Engineering, 2015; Volume 112, pp. 502-506. https://doi.org/10.1016/j.proeng.2015.07.232.
- Emad, Bassel & Atef, Omar & Shams, Yehya & El-Kerdany, Ahmed & Shorim, Nada & Nabil, Ayman & Atia, Ayman. (2020). iKarate: Karate Kata Guidance System. Procedia Computer Science. 175. 149-156. 10.1016/j.procs.2020.07.024.
- Kasiri, Soudeh & Fookes, Clinton & Sridharan, Sridha & Morgan, Stuart. (2017). Fine-grained action recognition of boxing punches from depth imagery. Computer Vision and Image Understanding. 159. 10.1016/j.cviu.2017.04.007.
- M. Zago, M. Codari, F. Marcello Iaia & C. Sforza Multi-segmental movements as a function of experience in karate, Journal of Sports Sciences, 2017; Volume 35, Issue 15, рр. 1515-1522, DOI: 10.1080/02640414.2016.1223332.
- Malawski, Filip & Kwolek, Bogdan. (2016). Classification of basic footwork in fencing using accelerometer. 51-55. 10.1109/SPA.2016.7763586.
- Messelodi, Stefano & Modena, Carla & Ropele, V. & Marcon, S. & Sgro, Michael. (2019). A Low-Cost Computer Vision System for Real-Time Tennis Analysis. 106-116. 10.1007/978-3-030-30642-7_10.
- N. Haralabidis, D.J. Saxby, C. Pizzolato, L. Needham, D. Cazzola, C. Minahan. Fusing Accelerometry with Videography to Monitor the Effect of Fatigue on Punching Performance in Elite Boxers. Sensors 2020; Volume 20, pp. 5749. https://doi.org/10.3390/s20205749.