Machine Learning Methods Used in the Automatic System for Teaching Human Motions—Key Aspects of CNN, HMM, and Minimum Distance Algorithms

Abstract

The article describes problems related to the construction of an automatic system for teaching human motor activities. Teaching these activities in rehabilitation, sports, and professional work is of great importance in both social and individual dimensions. The prospect of using automated systems is therefore highly significant. The system can use signals from any motion sensors, e.g., cameras or MEMS (Micro-Electro-Mechanical Systems) inertial sensors. A significant problem is real-time signal analysis. In the system presented, this analysis involves a classification process. It enables the selection of an optimal motor learning algorithm for a given situation. The learner is provided with information about required movement corrections through haptic devices. The primary aim of the research described in the article is to identify key features of classification methods that ensure the construction of an effective teaching system. To achieve this goal, three classification methods were statistically tested, namely: a method using CNN (Convolutional Neural Network), a minimum distance method, and a method based on hidden Markov models. The main result of the study is the statement that the key feature of the methods is their interpretability. This property enables the efficient transfer of knowledge from experts to the system and facilitates its improvement. Based on the research conducted, directions for further development of the system can be determined. The most important of them are developing software tools to help explain the work of minimum distance algorithms and creating a collection of CNN configurations for typical motor learning tasks.INDEX TERMS CNN networks, haptic feedback, machine learning, human-machine interface, MEMS sensors, motor learning, pattern recognition.