ИЕРАРХИЧЕСКАЯ МАШИНА СОСТОЯНИЙ ДЛЯ КЛАССИФИКАЦИИ ФИЗИЧЕСКИХ УПРАЖНЕНИЙ ПО ПОСЛЕДОВАТЕЛЬНОСТИ ПОЗ ЧЕЛОВЕКА

Abstract

In the last decade, significant resources have been devoted by international companies and research institutions to the development of neural networks for computer vision, which determine sequence of human poses from video. Since these data cannot be used directly by a human and requires pre-processing, therefore, it became necessary to develop universal methods for processing a sequence of human poses. The content and structure of the output signal after processing sequence of poses depends on the task, and in most cases are not universal. Versatile processing methods that can be used for different tasks are especially valuable. The article describes a method for processing of the output signal of a neural network, which allows you to determine the type of physical exercise. This method is quite universal and can be used independently or as one of the stages in solving a custom problem. One example of the application of the method is the automatic measurement of exercise duration during a sports session. Another example is the determination of the type of exercise in the case when this intermediate information is needed before applying the algorithms for counting the number of iterations of this exercise.