DEVELOPMENT OF KARATE KICK MODELS BASED ON VARIOUS TYPES OF ARTIFICIAL NEURAL NETWORKS

The work was aimed to develop punch recognition models based on a multilayer perceptron and convolutional neural networks. 
Methods and organization of the study. To study the kinematics of punch, inertial measurement modules (IMU) were used, with the help of which the accelerations of the punch segments of the body were measured. IMUs were fixed on the wrists of athletes. Studies were conducted for karate punches: gyaku-tsuki, mawashi-tsuki, age-tsuki, uraken. A class of movements without punches was also added. To develop punch models, the following deep learning methods were used: multilayer perceptron (MLP), 1-d and 2-d convolutional neural network (CNN). The F-score was used to evaluate the models. 
The results of the study. Studies have shown that good results were obtained for the MLP model, for example, the best result for the F-score is 0.95 for the uraken, the worst is 0.86 for the movement without punches. The difference between the accuracy of training and testing the developed model in the form of MLP may be the result of overfitting the model. The 1-d CNN model showed the worst results, apparently, it is poorly suited for recognizing bumps. The accuracy was about 0.8, and the accuracy of the test dataset was only 0.65 and was unstable. The worst result on the F-score is 0.11 for the mawashi-tsuki, the best result is 0.81 for the class without punches. The convolutional model with 2 layers performed better, approximately at the MLP level. The best indicator of the F-score was 0.93 for gyaku-tsuki, and the worst was 0.9 for uraken. 
Conclusion. A comparative analysis of the developed various models of artificial neural networks for recognizing punches in karate has shown that the multilayer perceptron is the simplest and most effective model.

• Cust E. E., Sweeting A. J., Ball K., Robertson S. Machine and deep learning for sport-specific movement recognition: a systematic review of model development and performance, Journal of Sports Sciences, 2019; Volume 37, Issue 5, pp. 568-600, DOI: 10.1080/02640414.2018.1521769.
• Dinu D., Millot B., Slawinski J., Louis J. An Examination of the Biomechanics of the Cross, Hook and Uppercut between Two Elite Boxing Groups. Proceedings, 2020, vol. 49, pp. 61, doi: 10.3390/proceedings2020049061.
• Filip M., Bogdan K. Recognition of action dynamics in fencing using multimodal cues, Image and Vision Computing, 2018, vol. 75, pp. 1-10, doi: 10.1016/j.imavis.2018.04.005.
• Haralabidis N., Saxby D.J., Pizzolato C., Needham L., Cazzola D., Minahan C. Fusing Accelerometry with Videography to Monitor the Effect of Fatigue on Punching Performance in Elite Boxers. Sensors 2020, vol. 20, pp. 5749, doi: 10.3390/s20205749.
• Howard R. Wireless Sensor Devices in Sports Performance. IEEE Potentials, vol. 35, no. 4, pp. 40-42, July-Aug. 2016, doi: 10.1109/MPOT.2015.2501679.
• Hülsmann F., Göpfert J., Hammer B., Kopp S., Botsch M. Classification of motor errors to provide real-time feedback for sports coaching in virtual reality. A case study in squats and Tai Chi pushes. Comput. Graph, 2018, vol. 76, pp. 47-59, doi: 10.1016/j.cag.2018.08.003.
• Kasiri S., Fookes C., Sridharan S., Morgan S. Fine-grained action recognition of boxing punches from depth imagery, Computer Vision and Image Understanding, 2017, vol. 159, pp. 143-153, doi: 10.1016/j.cviu.2017.04.007.
• Khasanshin I. Application of an Artificial Neural Network to Automate the Measurement of Kinematic Characteristics of Punches in Boxing. Appl. Sci, 2021; Volume 11, pp. 1223. https://doi.org/10.3390/app11031223.
• Kimm D. K., Thiel D. Hand Speed Measurements in Boxing. Procedia Engineering, 2015; Volume 112, pp. 502-506. https://doi.org/10.1016/j.proeng.2015.07.232.
• Wang Z., Yan W., Oates T. Time Series Classification from Scratch with Deep Neural Networks: A Strong Baseline. cs. LG, 2016, arXiv:1611.06455.
• Worsey M.T., Espinosa H.G., Shepherd J.B., Thiel D.V. An Evaluation of Wearable Inertial Sensor Configuration and Supervised Machine Learning Models for Automatic Punch Classification in Boxing. IoT, 2020; Volume 1, pp. 360–381, doi:10.3390/iot1020021.
• Worsey M.T., Espinosa H.G., Shepherd J.B., Thiel D.V. Inertial Sensors for Performance Analysis in Combat Sports: A Systematic Review. Sports (Basel), 2019; Volume 7(1), pp. 28. doi:10.3390/sports7010028.
• Zhang X., Chen X., Li Y., Lantz V., Wang K., Yang J. A Framework for Hand Gesture Recognition Based on Accelerometer and EMG Sensors. IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, vol. 41, no. 6, pp. 1064-1076, Nov. 2011, doi: 10.1109/TSMCA.2011.2116004.