NeuroSkate: neural network architecture for recognizing elements in figure skating

To achieve success in figure skating, hard work and effective interaction with the coach are important. In practice, the coach often fails to monitor the quantity and quality of the elements performed by young athletes in training. As a result, the aim of the work is to develop a neural network "NeuroSkate" designed to analyze the movements of child athletes in figure skating training. The work is being carried out in collaboration with the Orlyonok Olympic Reserve School in Perm, which acted as the main data provider for the study. The dataset includes a video of the performance of six key elements in figure skating by young athletes. The "NeuroSkate" architecture is a cascade of several architectures, each of which performs its own subtask: detection, allocation of key points and action classification. In the pairwise classification of performed elements, the proportion of a correctly defined class in the test sample of "NeuroSkate" is on average 72%.