Development of game module using technology of human pose estimation for the neurological rehabilitation system

The development of deep learning algorithms makes it possible to extend the scope of their application to various spheres of human life. Today, deep neural networks can solve problems in natural language processing, data generation, computer vision and so on. In this paper, a game module for a neurological rehabilitation system using Human pose estimation algorithm on video is designed and implemented. Different HPE algorithms including REMOTE, MAPN and MediaPipe Pose were considered in the research process and their comparative analysis on PCK, FPS and MAP metrics was done. As a result, MediaPipe Pose was selected to provide the best balance between accuracy and performance. The developed game module allows patients to perform movements in an interactive environment, and doctors to track rehabilitation progress based on movement parameters such as number of executions, time between executions, number of execution errors, and types of errors. The module allows doctors to select a difficulty level for the current game session to work with patients at different stages of rehabilitation.