Аналіз ланок тіла боксерів під час нанесення ударів за допомогою сучасних технологій

Abstract

Purpose: to evaluate the effectiveness of using modern computer vision technologies to determine the body parts of boxers during a punch. Material and methods. During the research, the following methods were used: analysis of scientific and methodological literature and Internet resources, analysis of videos and images using machine learning technologies (in particular, computer vision technologies), as well as the application of mathematical statistics methods. The final boxing match in the weight category up to 91 kg at the Olympic Games (Tokyo, 2021) was chosen for analysis. The total number of images for analysis was 1025. In the study, YOLO machine learning models were used to detect the persons who were in the images and MediaPipe to determine the body parts of each of the athletes. Results: as part of a scientific study of detecting the YOLO pattern during a boxing match, fairly high results were obtained. The first athlete in a red uniform was detected with high accuracy - he was not detected on only 1,4% of images; the second athlete in the blue uniform was detected in 98,7 % of all images. The probability of correctly identifying the athlete by the model was 89,5 % and 91,2 %, respectively, for the first and second athlete. Analysis of the results indicates that the MediaPipe model has limitations in identifying certain body parts of athletes during martial arts competitions. In particular, the percentage of images in which body parts could not be identified varies from 21,7 % to 31,7 %. The overall picture shows that the model successfully identifies key body parts such as shoulders, elbows, wrists, palms, and fingers, with a probability of results ranging from 61,5 % to 74,5 %. But visual direct analysis shows certain problems with determining the movements of athletes. Conclusions. An analysis of the results of the use of modern computer vision technologies to determine the body parts of boxers during striking in competitive activities was carried out. The results indicate the high efficiency and accuracy of the YOLO model in the task of detecting athletes during sports events. But at the same time, opposite data were obtained using the MediaPipe model to determine the body parts of athletes. Visual direct analysis shows certain problems with determining the movements of athletes. The general trend is that in the conditions of boxing competitions, the MediaPipe model may face challenges related to the specifics of this sport and require further optimizations to achieve the highest accuracy and reliability in the identification of boxers' body parts. But at the same time, the integration of computer vision technologies into sports events opens up new opportunities for objective analysis and improvement of the technical skills of martial artists. Keywords: boxing, competitive activity, impact, modern technologies, detection, modeling, body parts.