Análisis comparativo entre sistemas de clasificación basados en técnicas de Machine Learning para catalogar tipos de empuje de un patinador de velocidad

Abstract

Speed skating is a highly regarded sport that requires technical skills and an optimal physical condition; nevertheless, there is a need for a tool that complements the training of athletes in training. Thus, the use of artificial vision and Machine Learning (ML) systems is proposed. For the former, the OpenPose system is used to obtain data about the skater’s joins and keypoints; the latter, to identify and classify the type of push when skaters use the inline technique. In the research, 2 main focuses are explored. The first one consists of image classification through a Convolutional Neural Network (CNN), using the Visual Geometry Group (VGG)19 architecture, achieving an accuracy of 90.72 %. The second approach uses feature vectors through a biomechanical analysis with a Support Vector Machine (SVM) system and a Random Forest (RF) algorithm, obtaining an accuracy of 94 % y 92 %, respectively. Different evaluation metrics are presented, such as accuracy, precision, sensitivity, and F1-score, along with a confusion matrix and Receiver Operating Characteristic (ROC) curves. Moreover, features of the prior-mentioned approaches are discussed, for instance, the use of data augmentation to combat overfitting. Furthermore, the study includes result tables of biomechanical variables and pictures of distinctive angles during the execution of for- ward pushing techniques. Overall, the CNN and SVM exhibit promising results in the classification of data for the different types of pushes in speed skating inline technique.